arX

iv:m

ath-

ph/0

1010

02v1

2 J

an 2

001

Singularities, Structures and Scaling in Deformed Elastic m-Sheets

B.A. DiDonna and T.A. WittenDepartment of Physics, University of Chicago, Chicago, IL 60637

S.C. VenkataramaniDepartment of Mathematics, University of Chicago, Chicago, IL 60637

E.M. KramerDepartment of Natural Sciences and Mathematics,Simon’s Rock College, Great Barrington, MA 01230

The crumpling of a thin sheet can be understood as the condensation of elastic energy into anetwork of ridges which meet in vertices. Elastic energy condensation should occur in responseto compressive strain in elastic objects of any dimension greater than 1. We study elastic energycondensation numerically in 2-dimensional elastic sheets embedded in spatial dimensions 3 or 4 and3-dimensional elastic sheets embedded in spatial dimensions 4 and higher. We represent a sheet asa lattice of nodes with an appropriate energy functional to impart stretching and bending rigidity.Minimum energy configurations are found for several different sets of boundary conditions. Weobserve two distinct behaviors of local energy density fall-off away from singular points, which weidentify as cone scaling or ridge scaling. Using this analysis we demonstrate that there are markeddifferences in the forms of energy condensation depending on the embedding dimension.

PACS numbers: 68.60.Bs,02.40.Xx,62.20.Dc,46.25.-y

I. INTRODUCTION

In the last several years, there has been a marked interest in the nature of crumpling [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11].Field theories have been formulated for the crumpling transition [1], quantitative laws have been educed for the energyscaling of crumpled sheets [3, 4, 6], and dynamics of the crumpling process have been simulated and measured [7]. Inthis paper we treat crumpling as an example of energy condensation.The crumpling of a thin elastic sheet can be viewed as the condensation of elastic energy onto a network of point

vertices and folding ridges. These structures spontaneously emerge, for example, when a thin sheet of thickness h andspatial extent L ≫ h is confined within a ball of diameter X < L. For X ≤ L/2 the important length scales become hand X . The elastic energy scaling of vertices and ridges are well understood [3, 6]. In the limit h/X → 0, the elasticenergy is believed to condense into a vanishingly small area around the ridges and vertices.There is a significant body of physics literature on energy condensation, because it is a pervasive feature of condensed

matter. This behavior is seen in many systems including type-two superconductors [12], strongly turbulent flow [13] aswell as in mechanical [14] and electrical [15] material failure. Analogous condensation also occurs in particle-confininggauge field theories [16]. In a mathematical context, such condensation often arises in singular perturbations of non-convex variational problems [17, 18]. A few examples of such problems are the gradient theory of phase transitions[19], wherein the bulk of the energy is condensed into a small neighborhood of the interface between the two phases;Ginzburg-Landau vortices [20] which, among other things, describe type-two superconductors; and solid-solid phasetransitions in crystalline materials (martensitic phases) [17, 21].One distinctive aspect of the energy condensation in crumpling is the interesting dependence of the total energy

scaling on boundary conditions. A survey of the elastic energy scaling with thickness h for a given material with 2-dimensional strain modulus µ illustrates this point. For elastic sheets that are forced so that they form a single conicalvertex or “d-cone” [4], the only curvature singularity in the h → 0 limit is at the vertex of the cone [4, 9, 10]. The totalenergy of the sheet scales as µh2 log(X/h) [10] in this situation. When the boundary conditions are such that there aremany vertices and ridges (e.g. confinement), the elastic energy is concentrated on the ridges. For confined sheets, thetypical ridge length is on the order of the confining diameter X . It has been argued that ridges with length X have acharacteristic total elastic energy which scales as µh5/3X1/3 [22, 23], and that the total energy of the system scales withthe same exponent. A final example is the delamination and blistering of thin films, which is described by the sameenergy functional as the crumpled sheet but with different boundary conditions [24, 25, 26, 27, 28, 29, 30, 31, 32]. Inthis circumstance, the sheet develops a self-similar network of folding lines, whose lengths grow smaller as we approachthe boundary [31, 32], and the total energy of the sheet scales as µh (with a finite fraction of the energy concentratingin a narrow layer near the boundary, of a width that also scales as h [32]).Thus, by varying the boundary conditions, the same energy functional can lead to significantly different forms of

2

energy condensation, with different energy scalings and different types of energy bearing structures. This behavioris contrary to the widely held view that singularities are “local” phenomena. Our goal is to study this phenomenon,with a hope of understanding the factors that determine the nature of energy condensation in general systems. Inthis paper, we study elastic energy condensation in spatial dimensions above three. Our motivation is to understandhow the scaling behavior of crumpled sheets and the topology of energy condensation networks generalize for m-dimensional elastic manifolds in d-dimensional space. To this end, our numerical study explores energy condensationin 2-sheets in 3 or 4 dimensions and 3-sheets in dimensions 4 – 6, subject to boundary conditions which are akin toconfinement.Our previous work[33, 34] showed that the notion of an elastic membrane extends naturally to different dimen-

sions. Such membranes have an energy cost for “stretching” deformations that change distances between points inthe m-dimensional manifold and have an additional cost for bending into the embedding space. When these costsare isotropic, the material properties may be expressed in terms of a stretching modulus, a bending stiffness, anda“Poisson’s ratio” of order unity. As in 2-dimensional manifolds, the ratio of bending stiffness to stretching modulusyields a characteristic length. Indeed, if the manifold is a thin sheet of isotropic d-dimensional material, the thicknessh of the sheet is a numerical multiple of the square root of the modulus ratio that may be readily calculated[33].Our previous paper [35] identified two regimes of dimensionality with qualitatively different response to spatial

confinement. The authors considered an elastic m-dimensional ball of diameter L geometrically confined within d-spheres of diameter less than L/2. When the embedding dimension d is twice the manifold dimension m or more, thestate of lowest energy is one of non-singular curvature, with stretching elastic energy indefinitely smaller than bendingenergy. For the complementary cases where d is smaller than 2m, the deformation is qualitatively different. Suchmanifolds cannot be geometrically confined in a sphere of diameter smaller than L/2 without stretching or singularcurvature. In ordinary 2-sheets (m = 2) in 3 dimensions energy condenses in order to reduce the stretching energyof spatial confinement. The degree of energy condensation depends on the stretching moduli through the thickness hdefined above. In 3-sheets, singularities or stretching are required in 4 or 5 embedding dimensions. Previous work [34]confirmed that for 3-sheets confined in 4 dimensions, energy condenses into a network of line-like vertices and planarridges. We seek to understand how the degree of energy condensation associated with confinement changes withincreasing spatial dimension. We expect that less energy will be required to confine a 3-sheet within a 5-dimensionalsphere than within a 4-dimensional sphere, but we do not know a priori how the form of energy condensation willdiffer between these two cases.We begin our study by giving a brief review of elastic theory in Section II. Then, Section III quantifies our definitions

of “folding lines” and “vertices” within a framework of isometric embeddings, and in Section III A we propose a rulefor the topological dimensionality of vertices in energy condensation networks. In Section IV we present analyticalestimates for the degree of energy condensation in the crumpled state. Building on existing knowledge, we makepredictions for the scaling of energy density with distance away from the regions of greatest elastic energy. Weidentify two distinct forms of energy scaling, which we call ridge scaling and cone scaling (the names are based on thegeometry these scalings correspond to in ordinary crumpling of 2-sheet in 3 dimensions). Section V describes how werepresent elastic manifolds numerically.Then we present our numerical findings. We begin in Section VI with simulations of sheets confined by shrinking

hard wall potentials. In this qualitative study the embedding dimension seems to affect the crumpled structure signif-icantly. The condensation of energy appears to become progressively weaker as the embedding dimension is increased,culminating in no condensation when d reaches 2m. Numerical difficulties prevented any significant quantitativeanalysis of the geometrical confinement data. The need for better data motivates the simpler systems we simulatednext.Section VII describes our studies of m-sheets with two disclinations. Disclination are made by removing wedge

shaped sectors from the sheet and then joining the edges of each wedge. The essential feature of a disclination isthat it induces the sheet to form a cone, with lines of null curvature converging at a vertex. It has been shown thatwhen two disclinations are introduced into a 2-sheet in 3 dimensions, the elastic energy of deformation between thedisclinations condenses along a ridge joining the two vertices [3]. These ridges appear completely similar to those ingeometrically confined sheets and exhibit the same energy scaling [22]. In our present study, simulated 2-sheets in 3dimensions formed the familiar ridges, but 2-sheets with the same boundary conditions in 4-dimensional space hadmuch lower total elastic energies and very different energy distributions. Similarly, 3-sheets in 4 spatial dimensionsformed ridges closely analogous to those seen in 2-sheets, but for 3-sheets in 5 dimensions no ridges were evident. Also,non-parallel disclination lines in 3-sheets appear to generate further disclination-like lines in 4 spatial dimensions butnot in 5.Next, in Section VIII we detail our simulations of 3-tori allowed to relax in d dimensions. The benefit of this

geometry is that we expect it to cause energy condensation without the need to introduce disclinations. Observingthat a 2-torus cannot be smoothly and isometrically embedded in a space of dimensionality less than 4, we expect anelastic sheet with the connectivity of an m-torus embedded in a space of dimension d < 2m will relax to a configuration

3

with regions of non-zero strain (condensed into a network of ridges). We found that a 3-torus in d = 4 spontaneouslyforms a network of planar ridges which intersect in vertex singularities similar to those in the geometrically confinedsheets. In d = 5, the 3-torus forms a point-like vertex network with no observable ridges. The energy scaling andpresence or absence of ridges mirrored the behavior of sheets with disclinations in Section VII. The complexity of thecrumpling network decreases with increasing embedding dimension, with spontaneous symmetry breaking evident ford = 5. As expected, the elastic energy distribution is homogeneous for d ≥ 6.Finally, in Section IX we present the results of simulations of a “bow configuration”, in which the center points

of opposite faces of a 3-cube were attached and the cube was embedded in 4 or 5 spatial dimensions. With propermanipulation of initial conditions, the cube embedded in 5 dimensions forms a single, point-like vertex at its center.The energy density scaling away from this singularity agrees with predictions for a novel kind of elastic structurewhich is a generalization of a simple cone. By contrast, the cube embedded in 4 dimensions forms a set of line-likevertices and planar ridges that are well modeled by our present understanding of 3-dimensional crumpling.We conclude by discussing the observed energy scaling properties of crumpled elastic sheets. We have developed a

means to identify the presence of ridges in m-sheets based solely on their spatial elastic energy distribution. Usingthe analysis of energy distributions, we demonstrate that folding lines in greater than m+1 dimensions have differentenergy and thickness scaling properties than in m+1, but ridges in m+1 seem to have the same scaling regardless ofm. We found that ridge scaling dominates the crumpling of m-sheets in m+1 dimensions, while cone scaling was theonly form of scaling witnessed in dimensions greater than m+1. Differences in the morphology of higher dimensionalfolding lines is discussed briefly. The local structure of these folds is very different from that of the familiar ridgesfound in 2-sheets in 3-dimensional space. We also note that our simulational findings strongly support the new rulefor the topology of elastic energy bearing structures in higher dimensions which is presented in section III A. We endwith a brief discussion of the mathematical questions raised by the non-local character of energy scaling in crumpledsheets.

II. ELASTIC m-SHEETS IN d-SPACE

In this section, we review the elastic theory of m-sheets in d-space as it is presented in Ref. [33]. In analogy withthe elastic 2-sheets of everyday experience, an m-sheet in d dimensional space is an elastically isotropic d-dimensionalsolid which has a spatial extent of order L in m independent directions and h ≪ L in the remaining d−m directions.Specifically, our m-sheet is given by S × Bd−m

h ⊂ Rd, where S ⊂ R

m is a set that has a typical linear size L in all

directions, and Bd−mh is a d−m dimensional ball of diameter h.

We are considering embeddings of the m-sheet in a d-dimensional target space. We first consider the lowestenergy embedding in a sufficiently large d-dimensional space, say all of Rd, so that the sheet is not distorted in theembedding. We assume that the undistorted sheet has no intrinsic strains, curvatures or torsions(twists). Since thereare no curvatures or torsions, picking a orthogonal basis of d −m vectors for the thin directions at one point on thesheet, and then parallel transporting these vectors to every point on the sheet gives an orthonormal set of basis vectorsfor every point of the sheet. We can therefore describe the geometry of the undistorted sheet, which is a d-dimensionalobject, by the m-dimensional center surface S which gives the geometry in the long directions, and the orthonormalbasis vectors for the thin directions, that describes the geometry in the thin directions. These basis vectors for thethin directions give a normal frame field to the embedding of the center surface, since they are all orthogonal to eachof the long directions in the sheet. Further, since the basis vectors at different points are related by parallel transportin R

d, the normal frame field is torsion-free.For small distortions, we can continue to describe the embedding of the m-sheet by giving the embedding of the

center surface S and by specifying the normal frame field [33]. The rotational invariance in the thin directionsimplies that the torsion of the sheet in the embedding cannot couple to the geometry of the center surface. Sincethe sheet has no intrinsic torsion, if there are no applied applied torsional forces, the normal frame has to remaintorsion free. The torsion degrees of freedom therefore drop out of the energetic considerations that will determine thegeometry of the sheet [33]. Therefore, we can leave out the thin directions and determine the energy of the embeddingthrough an effective Lagrangian that only depends on the long directions, i.e., the geometry of center surface ofthe sheet S, as embedded in the d-space [33]. This approach puts powerful tools of differential geometry at ourdisposal. Numerically, this treatment greatly increases the efficiency of our simulations by decreasing the dimensionand required grid resolution of our lattice. In the limit h/L ≪ 1 and for relatively small elastic distortions of thematerial, this description is highly accurate.We use Cartesian coordinates in the center surface, which can be viewed as the set S ⊂ R

m. We refer to thesecoordinates as the material coordinates, and quantities referred to the material co-ordinates will be denoted by Romansubscripts e.g. i, j, k, l. The configuration of the sheet is given by a vector valued functions ~r(xi) with values in thed-dimensional target space. We also denote the d−m normal vectors in a choice for an orthonormal, torsion-free frame

4

by ~n(α), with a Greek superscript that takes values 1, 2, . . . , d−m. Such a choice exists by our previous considerations.The strain energy density Ls due to the distortions within the m-sheet is given by the conventional expression [36]

in terms of the Lame coefficients λ and µ

Ls = µγ2ij +

λ

2γ2ii, (1)

where γij is the strain tensor, defined by

γij =1

2

(

∂~r

∂xi· ∂~r

∂xj− δij

)

.

The strain tensor quantifies the deviation of the metric tensor of the embedded sheet from it’s intrinsic metric tensor.Here and henceforth, repeated indices (both Greek and Roman) are summed over all the range of their allowed values.The non-zero thickness of the m-sheet leads to an energy cost for distortions of the center surface S in a normal

direction, i.e., bending distortions. A measure of the bending of the manifold at any point is the extrinsic curvaturetensor ~κij(xi), which is the projection of the second derivatives ∂i∂j~r into the normal frame. The component of the

extrinsic curvature in the normal direction ~n(α) is given by

κ(α)ij =

∂2~r

∂xi∂xj· ~n(α). (2)

As shown in Ref. [33], if the strains are small and the curvatures are small compared to 1/h, the energy density ofthe bending distortions Lb is given by

Lb = B

[

κ(α)ij κ

(α)ij +

λ

2µκ(α)ii κ

(α)jj

]

. (3)

The bending modulus B in the above equation is determined by the Lame coefficient µ and the thickness of the sheeth through the relation [33]

B = µh2/η(m, d),

where η(m, d) is given by

η(m, d) =d−m

Sd−m×

23 d−m = 1π4 d−m = 2

1d−m+2β(3/2, d−m− 2)Sd−m−1 d−m > 2

, (4)

where Sa = 2πa/2/Γ(a/2) is the area of a unit sphere in a dimensions and β(a, b) = Γ(a)Γ(b)/Γ(a + b) is the betafunction. For m = 3 and d = 4, 5, 6, η(m, d) = 3, 4 and 5 respectively.For studying the geometrical confinement of an elastic m-sheet, the confining forces are assumed to be derived from

a potential Vc(~r) in the embedding space. The energy of the m-sheet is the sum of the bending energy, the strainenergy and the energy due to the spatially confining potential. Therefore, the total energy is given in terms of thegeometry of the center surface S by

E = µ

∫

S

dmx

[

h2

η

(

κ(α)ij κ

(α)ij +

λ

2µκ(α)ii κ

(α)jj

)

+

(

γ2ij +

λ

2µγ2ii

)

+Vc(~r(xi))

µ

]

, (5)

where η = η(m, d) as defined in Eq. 4.The configuration ~r(xi) of the sheet in the embedding space is obtained by minimizing the energy E over the set

of all allowed configurations. A (local) minimum energy configuration is obtained by requiring that the variation δEshould vanish to the first order for an arbitrary (small) variation δ~r of the configuration. Since the energy densitycontains terms in κij , that involve the second derivatives of the function ~r(xi), the Euler-Lagrange equations forthe minimization problem are a system of fourth order, nonlinear elliptic equations on the domain S. Very little isknown about the rigorous analysis of such equations. Therefore, we will study the geometrical confinement problemnumerically, by approximating the the integral in Eq. 5 by a sum over a grid, and minimizing the resulting energy bya conjugate gradient method [37], as we outline below.Our goal is to study the scaling behavior of the structures on which the energy concentrates as the thickness h → 0.

The variational derivative of the potential term is given by

δ

δ~r

∫

S

dmxVc(~r(xi)) = ∇~rVc(~r(xi)).

5

This term leads to a strongly non-linear coupling between the configuration of the minimizer and the stresses andthe bending moments in the sheet. Consequently, the conditions for mechanical equilibrium are now “global” and thestresses and bending moments determined by the local strains and curvatures should balance a term that dependson the global geometry of the configuration. In addition to complicating the analysis, this introduces length scalesbesides the thickness h into the problem. This in turn can lead to the lack of simple scaling behavior at equilibriumfor the structures in geometrically confined sheets. Note however, that this is not the case for confinement in a hardwall potential

Vc(~r) =

V0 for ~r ∈ Ω+∞ otherwise

where Ω is a given set in Rd. The configuration of the minimizer is now restricted to be inside Ω and the gradient of

Vc is zero here, so that there is no coupling between the the configuration of the minimizer and the stresses and thebending moments in the sheet, in the parts of the sheet that are in the interior of Ω.One way to get around this problem is to study the configurations where the energy concentration is due to the

boundary conditions imposed on the sheet, and not due to an external potential. If the imposed boundary conditionsdo not introduce any new length scales, we would then expect to see structures and scalings that are generic, i.e.,independent of the precise form of the imposed boundary conditions. This is analogous to the minimal ridge [22] thatis obtained by imposing boundary conditions on a 2-sheet. Although the minimal ridge is obtained with a specificboundary condition, the scaling behaviors of the ridge are generic and are seen with a variety of boundary conditions.In this study we first determine the generic structures and scalings that we expect to see for an m-sheet in d-

dimensional space. We also numerically verify our predictions for these scalings by the configuration of an embeddedm-sheet with a variety of boundary conditions – sheets with disclinations, sheets with a toroidal global connectivity,and sheets in a “bow” configuration. In all these cases, the elastic energy is given by

E = µ

∫

S

dmx

[

h2

η

(

κ(α)ij κ

(α)ij +

λ

2µκ(α)ii κ

(α)jj

)

+

(

γ2ij +

λ

2µγ2ii

)]

. (6)

However, the domain of integration S is no longer a subset of Rm. It is a domain with singularities in the case ofsheets with disclinations or in the “bow” configuration, or a set whose global topology is different from R

m, in thecase of the sheets with toroidal connectivity. Note that this energy functional is also applicable to the confinementin a hard wall potential, since, without loss of generality, we can set V0 = 0 for ~r in Ω, and impose the constraint ofthe hard wall potential through the conditions ~r(xi) ∈ Ω for all xi ∈ S. Consequently, the energy is still given byEq. 6, and the energy condensation is due to the additional constraints that are imposed, that are analogous to theboundary conditions considered above.

We can rewrite the energy using the in-plane stresses σij and the bending moments M(α)ij that are conjugate to the

strains γij and the curvatures κ(α)ij respectively. The conjugate fields are given by the variational derivatives

σij =δEδγij

= 2µγij + λδijγkk,

and

M(α)ij =

δEδκ

(α)ij

=h2

η

(

2µκ(α)ij + λδijκ

(α)kk

)

,

where we have taken the variational derivatives as though the fields γij and κ(α)ij are independent. The energy can

now be written as

E =1

2

∫

S

(

M(α)ij κ

(α)ij + σijγij

)

dmx.

Although the energy functional E does not explicitly couple the strains in the manifolds to the curvatures (See Eq. 6),they are related by geometric constraints since they are both defined by derivatives of the embedding ~r(xi). Since them-sheet is intrinsically flat, the Riemann curvature tensor for the embedding of the center surface can be expressed

in terms of the extrinsic curvature κ(α)ij by the Gauss Equation [38],

Rijkl[κ] = κ(α)ik κ

(α)jl − κ

(α)il κ

(α)jk .

6

However, the Riemann curvature is intrinsic to the geometry of the center surface, and can be written in terms of thestrains as

Rijkl [γ] = −γik,jl + γil,jk − γjl,ik + γjk,il +O(γ2).

Consequently, the curvatures κ(α)ij and the strains γij are constrained in order that Rijkl [κ] = Rijkl[γ].

From the symmetries of the Riemann tensor, it has m(m− 1)(m2 −m+ 2)/8 independent components. However,since it can be written purely as a function of the strain γij , it can only have as many independent degrees of freedomas the strain itself. As noted in Ref. [33], the strain tensor is symmetric, and further it satisfiesm additional conditionsfrom the balance of in-plane stresses. Consequently, the strain has m(m − 1)/2 independent components, and thisyields m(m− 1)/2 independent constraints on the extrinsic curvatures.For m = 2, i.e., for 2-sheets, there is one constraint, and this is most economically expressed through the Gaussian

curvature of the sheet [33]. In terms of the extrinsic curvatures, the Gaussian curvature G is given by

G[κ] = κ(α)11 κ

(α)22 − κ

(α)12 κ

(α)12 ,

and in terms of the strains, the Gaussian curvature is given by

G[γ] = −γ11,22 + 2γ12,12 − γ22,11 +O(γ2).

We can impose the constraint G[κ] = G[γ] through a Lagrange multiplier χ, so that the augmented energy functionalis now given by [33]

Eχ =

∫

S

dmx

[

1

2

(

M(α)ij κ

(α)ij + σijγij

)

+ χ(G[γ]−G[κ])

]

.

Taking the variations with respect to γij , χ and κ(α)ij give

σij = δij∇2χ− ∂i∂jχ ⇒ ∂iσij = 0,

G[γ] = G[κ],

and ∂i∂jM(α)ij = σijκ

(α)ij ,

which are respectively, the balance of the in-plane stresses, the Geometric (or First) von Karman equation and theForce (or Second) von Karman equation [33, 39]. The first equation also shows that the Lagrange multiplier χ is thescalar stress function of Airy [40].In this work, we will mainly focus on the case m > 2. For m > 2, an economical way to impose the geometric

constraint relating the extrinsic curvatures to the strains is through the Einstein tensor [33]

Gij = Rikjk − 1

2δijRlklk,

which has m(m−1)/2 independent components since it is symmetric and satisfies the contracted Bianchi identity [41]

∂iGij = 0.

In terms of the extrinsic curvature,

Gij [κ] = κ(α)ij κ

(α)kk − κ

(α)ik κ

(α)jk − 1

2δij

[

κ(α)ll κ

(α)kk − κ

(α)lk κ

(α)lk

]

, (7)

and to the first order in the strains

Gij [γ] = −γij,kk + γik,jk − γkk,ij + γkj,ik + δij [γll,kk − γlk,lk] . (8)

As in the case m = 2, the constraint Gij [κ] = Gij [γ] is incorporated through a tensor Lagrange multiplier χij . Theaugmented energy functional is given by

Eχ =

∫

S

dmx

[

1

2

(

M(α)ij κ

(α)ij + σijγij

)

+ χij(Gij [γ]−Gij [κ])

]

.

7

Taking the variations with respect to γij , χij and κ(α)ij give the balance of in-plane stresses

∂iσij = 0,

the Geometric von Karman equation

Gij [γ] = Gij [κ], (9)

and the Force von-Karman equation

∂i∂jM(α)ij = σijκ

(α)ij , (10)

respectively [33]. In the case m = 3, the Lagrange multiplier χij is the Maxwell stress function [42].

III. STRUCTURES IN ELASTIC m-SHEETS

We will now investigate the minimum energy configurations of the sheet with external forcing. As we discussedearlier, the sheet can be forced either by an external potential Vc(~r) (See Eq. 5) or by restricitng the set of admissibleconfigurations by appropriate boundary conditions (Eq. 6). Since confinement by a hard wall potential of radius r0is also given by the energy functional in Eq. 6 where the admissibility condition is that ‖~r(x)‖ ≤ r0 for all x ∈ S, wewill restrict our attention to the energy functional E in Eq. 6.From Eq. 6, we see that the only length scales in the energy functional are the thickness h and the length scale

L that is associated with the center surface S. Since the effective bending modulus µh2/η goes to zero as h → 0,except in the vicinity of regions with large curvature, the large scale (O(L)) behavior of crumpled sheets should bedetermined almost entirely by the stretching energy functional

Es = µ

∫

S

(

γ2ij + c0γ

2ii

)

dmx,

which penalizes the deviation of the configuration from an isometry. Indeed, crumpled 2-sheets in 3 dimensions can bedescribed as a set of nearly isometric regions bounded by areas of large curvatures that include vertices and boundarylayers around folds. Since the curvature in these regions is large, the bending energy in this region will continue toremain relevant as h → 0. As h → 0, the width of, and the strain in, the boundary layer around folds goes to zero,and the only non-isometric regions are the vertices.For the remainder of this paper we assume that in any dimension, the minimum energy configurations of crumpled

m-sheets converge in the h → 0 limit to configurations which are locally isometric and have smooth, well definedcurvature almost everywhere. In this view, the regions of elastic energy concentration in the m-sheets converge in theh → 0 limit to a defect set in the manifold which is not locally smooth and isometric, and this defect set is as small aspossible relative to the boundary conditions imposed on the sheet. The limiting procedure which connects the defectset to the energy concentration regions is elaborated upon in Sec. III A.These assumptions give us descriptive tools to classify the elastic energy structures in higher dimensional crumpled

sheets in terms of well defined concepts of isometry. More importantly, the identification of crumpling with isometricembedding will allow us to make predictions for the dimensionality of energy condensation regions in m-sheets form > 2 based on geometric results on isometric immersions. We will present these arguments in Sec. III A and thenumerical studies reported in Sections VII - IX appear to support these predictions.

A. Dimensionality of Defects

In this section we define a certain type of singularity called a vertex that must exist in confined m-sheets. We thenargue that a vertex must have a dimensionality of at least 2m−d−1. In previous work [35] we showed that a m-sheetembedded smoothly and isometrically into a space of dimension less than 2m must have straight lines in the sheetmaterial which extend across the sheet and which remain undeformed. Specifically, there exists through any point pat least one straight line in the undistorted m-sheet S which 1) is straight and geodesic in the embedding space R

d,and 2) extends to the boundary of S. We’ll denote this result as Theorem 1. Theorem 1 implies that a m-sheet ofminimum diameter L cannot be confined to a d dimensional ball of radius smaller than L/2, if d < 2m, where theminimum diameter L is given by

L = 2maxp∈S

(maxr>0

r : B(p, r) ⊆ S),

8

FIG. 1: Flat subspaces R in a crumpled sheet. The figure shows sections of three parallel planes of a thin 3-cube confinedin four dimensions by boundary conditions discussed in Section IX. This sheet is nearly isometric over most of its volume, asanticipated in Section IIIA. The arguments of this section suggest that such sheets should have a nearly-flat 2-dimensionalsubsheet R through any point p. The R for the indicated point p is shown as a solid grid. The nearly-flat subsheet R′ for adifferent point p′ is shown as a dashed grid. The adjacent boundaries of R and R′ meet in a nearly-straight, 1-dimensionalregion. We identify this region as a vertex.

and B(p, r) is the m dimensional ball of radius r centered on p. By taking points far from the boundary of S, we mayidentify lines roughly of the size L/2 or longer. It is clearly impossible to confine the sheet to a region smaller thansuch a line.Observations of embedded sheets and generalizing the proof from [35] lead us to conjecture the following extension

to Theorem 1. We conjecture that through any point p there is a (2m− d)-dimensional subsheet [46]

1. R is totally geodesic in the sheet.

2. The image of R under the embedding is totally geodesic in Rd. This together with item 1 implies that the sheet

is flat.

3. If the point p is a distance X from the boundary of S, then the subsheet R through p contains a (2m − d)-dimensional ball of diameter X .

We’ll denote these assertions as Conjecture 1. The subsheets R can be readily be identified for a simple cone in atwo sheet. For any point p on the cone, R is the half-line extending from the apex through p. Figure 1 illustrates anexample of a 3-sheet in which the subsheets R are two-dimensional.We may in principle confine an m-sheet isometrically within a ball of arbitrarily small size by removing subsets of

S so that it has “interior” boundaries. We shall denote the removed part as the defect set D. By removing sufficientlymany subsets, we can assure that all points of the resulting sheet S ′ are as close to the (interior) boundary as welike. In order that the remaining region be isometric, further conditions are needed: Conjecture 1 forces some of theremoved regions to have a dimensionality greater than some limit, as we now show.We first confine a convex m-sheet S within a ball of diameter X much smaller than the minimum diameter L of

the sheet. As indicated above, this confinement requires strain or singularities. We now remove a defect set D fromthe sheet sufficient to allow the remaining sheet S ′ to be isometric, as illustrated in Figure 2. We choose a point pfurther than X from the original S boundary, as measured along the sheet. The subsheet R at point p can have aminimum diameter no greater than X ; otherwise this flat subspace would not fit into the confining ball. Thus theoriginal boundary of S cannot touch the boundary of R; R must be bounded everywhere by D. Now, since R is a(2m − d)-dimensional set, at least part of its boundary must have dimension at least (2m − d − 1). (The boundarymay also have additional parts of lower dimension, but we ignore these.) The set D adjacent to this boundary musthave at least this dimension as well. Thus, most R’s in the sheet must be bounded over part of their boundary bydefect sets D whose dimension is (2m− d− 1) or more.

9

(a) (b)

δ δ

δ

FIG. 2: (a) Illustration of the regions D, K, Dδ, and Kδ for a 2-sheet. The points are a possible set D and the shaded circlesare the corresponding Dδ. The solid lines are a possible set K−D, and the area within the dashed lines are the correspondingKδ −Dδ . (b) Illustration of a potential way to soften the folding around a region in K.

These defect sets in strictly isometric sheets have implications for the confinement of real elastic sheets. To see this,we repeat the confinement procedure above taking S to be an elastic sheet of thickness h. We anticipate that regionsof concentrated strain will appear, as they do in ordinary crumpled 2-sheets. Following the procedure used above, weremove part of S near the regions of greatest strain, such as the intersection of R and R′ in Figure 1. Specifically, weremove sets of minimum diameter δ, and denote the set of removed points Dδ. We remove the smallest set such thatthe remaining sheet S ′

δ becomes isometric in the limit as h → 0. We now reduce the minimun diameter δ of our setand repeat the procedure. We suppose that the new defect set Dδ is a subset of the old one, and that we are led to awell-defined limiting set D as δ → 0. For each δ we may consider the boundary of R for a given point p. Supposingthat this boundary also behaves smoothly, we infer that it retains its dimensionality of at least 2m − d − 1 inferredabove. Thus the limiting defect set D should also have at least this dimension. Returning now to the full elastic sheetS, we expect the strain to be concentrated on the defect set D. The example of Figure 1 suggests that R sets arebounded by regions of high strain, whose dimension has the minimal value 2m− d− 1. The numerical work in latersections gives more systematic evidence of these strained regions. We shall denote the limiting set D as the strain

defect set and denote each connected part of D as a vertex.Although the elastic sheet S ′

δ becomes isometric as h → 0, further singularities can develop as δ → 0. Ordinary2-sheets in 3-space show this behavior, as illustrated in Figure 2. Here the minimal vertex dimension 2m− d− 1 is 0.The set Dδ consists of the four shaded disks: each disk constitutes a vertex. Removing these disks permits strain-freeconfinement to a fraction of the size of the sheet. However, the strain-free deformation develops large curvature asδ becomes small. The diverging curvature is concentrated on lines joining the vertices. Similar diverging curvaturemust occur in intact sheets as h → 0. We denote such regions by K, which we call the curvature defect set. Forcompleteness, we define a set Kδ which contains the regions of high curvature around K for δ > 0. For intact sheets,we expect the strain to be significant in the region D but very small outside of it – noting that the geometric vonKarman equation, Eq. 9, relates large gradients in the strain to large curvature, we conclude that D must be a subsetof K. We denote each connected piece of K−D as a fold in the crumpled sheet. The relationship between these foldsand stretching ridges [23] is discussed in the next section.Thus far we have considered effects due to confinement in a small ball in R

d. We expect similar effects if we imposeother constraints that reduce the spatial extent of the embedded sheet. We expect defect sets D and K like thoseabove to form spontaneously here as well. Our numerical investigations reported below do indeed show such behavior.We compare our expectations with the numerical findings in Section X.

IV. ENERGY SCALING

We now return to the consideration of sheets with small h > 0. We consider sheets which are thin enough thatthe strains far away from the singular set are much less than O(1). This is the range of thicknesses that is normallyconsidered in the study of thin 2-sheets [6, 22]. In this range the preferred configuration of a 2-sheet is well describedby asymptotically matching nearly isometric embedding over most of the sheet to finite boundary layers around thesingular set K (within which the strains and curvatures may become large on the scale of L, the manifold size). Wemaintain the assumption that energetically preferred embeddings will exhibit near isometry outside the singular setfor m-sheets in d-dimensional space, and view the singular set as the subset of the material manifold onto which

10

(a) (b)

(c) (d)

FIG. 3: A cone and a ridge formed with disclinations. Image (a) shows a minimal elastic energy embedding for a 2-sheet withone disclination in 3-dimensional space. The sheet was 50× 100 lattice units in size and had an elastic thickness of ∼ 1/100 inlattice units. The disclination was formed by folding one edge at its center point and attaching the two halves. The minimalenergy configuration is a cone. Plot (b) shows surfaces of equal bending energy for the sheet in (a), plotted in its materialcoordinate system. Image (c) shows a minimal elastic energy embedding for a 2-sheet with two disclinations in 3-dimensionalspace. The sheet was 100× 100 lattice units in size and had an elastic thickness of ∼ 1/10 in lattice units. The minimal energyconfiguration is a ridge. Plot (d) shows surfaces of equal bending energy for the sheet in (c), plotted in its material coordinatesystem. In each image, heavy lines indicate edges of the sheet which were joined together to make the disclinations.

elastic energy condenses as h → 0. We wish to study the degree of energy condensation onto these elastic structuresas a function of the material and embedding dimensions and the elastic thickness h. In this section, we show how thescaling of elastic energy density with volume away from the condensation regions can be used to quantify the degree ofenergy condensation in crumpled m-sheets. We distinguish two cases for the structures involved in crumpling. In thefirst case, K = D and singular curvature occurs only at vertices. In the complimentary case, K−D 6= ∅, vertices areconnected by folds in the h → 0 limit. We show that these two cases have distictive energy scaling signatures when2-sheets are crumpled in 3 dimensions. Anticipated scaling exponents for general crumpling are inferred by analogyto lower dimensional crumpling.There are three types of data we may use to analyse minimum energy sheet configurations: the detailed embedding

coordinates and the bending and stretching energy densities in the manifold coordinates. To see whether energyhas condensed in our simulated sheets, we first identify regions of high energy concentration by plotting surfaces ofconstant bending or stretching elastic energy in the material coordinates. Fig. 3 illustrates, for the case of a 2-sheetin 3 dimensions, how surfaces of constant bending energy highlight the energy-bearing regions of the sheet. We thenlook at the coordinate information to associate regions of energy concentration with either vertices or folds.For the remainder of our analysis we consider the energy density data only as a function of volume fraction,

independent of position. This removes any ambiguity in defining the center points of the high strain regions aroundD, and it provides a natural framework for defining the degree of energy condensation. Let the variable Φ represent thevolume fraction in the manifold coordinates measured from the regions of highest to lowest energy density, 0 ≤ Φ ≤ 1.Thus, Φ can be written:

Φ (L) = 1

Vtotal

∫

L(~x′)≥L

dmx′, (11)

11

where L = Lb + Ls is the elastic energy density defined in Eqs 1 and 3, and

Vtotal =

∫

S

dmx′.

Surfaces of constant energy in the manifold coordinates are also surfaces of constant Φ. Inverting Eq. 11 associates avolume fraction with each observed value of the local elastic energy density. We can write the total energy E in themanifold as

E =

∫ 1

0

dΦ′L(Φ′). (12)

We say the energy is condensed in a volume fraction Φc if for Φ > Φc the energy density L falls away faster than Φ−1.If this is the case, then the upper limit of integration can be pushed to infinity without changing the value of theintegral by more than a finite fraction. By repeating this analysis for the bending energy Lb or the stretching energyLs, we may characterize the condensation of these forms of energy individuallyWe may now make predictions for the elastic energy scaling exponents based on our knowledge of the structures

found in crumpled sheets. We first consider the case when the set K = D. In the familiar crumpling of 2-sheets in 3dimensions K = D when the sheet contains a single vertex and the configuration outside the vertex is conical. In anydimension, it is easy to see that far away from D the conformation should be independent of the small length scale h.Far away from the curvature singularity there is no intrinsic length scale, so simple dimensional analysis tells us thatthe curvature must be a numerical multiple of r−1, where r is the distance from the vertex. The preferred embeddingis thus a simple generalization of a cone, with straight line generators radiating from a central vertex and transversecurvatures decreasing as 1/r along the generators. The cone configuration is isometric outside the vertex for h = 0,but for h > 0 it acquires small but finite strain. Dimensional analysis of the force von Karman equation, Eq. 10,for a curvature of the form C(r, θ) = g(θ)/r yields strain scaling of the form h2/r2 for nearly isometric embeddings.Thus, for energetically preferred embeddings, the bending and stretching energy densities should scale as 1/r2 andh2/r4 respectively. We can express this energy scaling in terms of the volume fraction Φ by finding how Φ growswith r. In any principal material direction, smooth curvature of order C will typically persist over a length of order1/C. In a 3-sheet, the volume of high energy density surrounding an energetic cone generator will therefore grow asr2 if the curvature in one material directions transverse to the generator dominates, or as r3 if the curvatures in bothtransverse directions are on the same order. The bending energy density will respectively scale as Φ−1 or Φ−2/3. Sincethe strain along a generator dies twice as quickly as the curvature, the stretching energy density will correspondinglyscale as Φ−2 or Φ−4/3. Thus we surmise that conical scaling has the typical form

Lb ∼ Φ−p

Ls ∼ Φ−2p,(13)

where p = 2/(1 + n), and n is an integer equal to the number of transverse curvature directions along energetic conegenerators. In all the geometries accessible to 2 or 3-sheets, the stretching energy is condensed while the bendingis not – this is not the case in all higher dimensions, where the value of n can be greater than 3 and the stretchingenergy not condensed.We now consider configurations that have K − D 6= ∅. We have denoted each connected piece of K − D as a fold

in the previous section. At this point we need to make a distinction between folds and ridges. For 2-sheets in 3dimensions with h > 0, folds have an energetically preferred local structure. We describe folds in this context asridges, a term which encompasses both the geometric and energetic structure. In general crumpling, we don’t knowa priori whether the local structure around folds will be similar to that in lower dimensional crumpling, so we mustmake our definition of a ridge more precise. Since we already have a geometrical descriptor for folds, we use theterm ridge to describe a certain kind of energetic structure associated with folds. The generalization of a ridge is aboundary layer around a fold whose energy scaling depends on two length scales — the elastic thickness h of the sheetand the length L of the fold. Clearly in the thin limit these are the only two length scales which can be importantaround the fold. Conversely, there must be at least two length scales if there is to be any non-trivial scaling of theridge profile with thickness h. The presence of two length scales allows for a balance between the coupled bendingand stretching energies, which is also a hallmark of ridges (and could be used as an alternative equivalent definition)For 2-sheets in 3 dimensions, the condition K − D 6= ∅ occurs e.g. when there are two vertices joined by a ridge,

as in Figure 3. It is well known [22] that the elastic energy density in the region surrounding K which encompasses aridge is less than that at vertices (the region surrounding D) but much greater than that in the region of S −K awayfrom the energy condensation structures. Ridges begin and end at vertices, with the elastic energy density fallingsmoothly along the ridge length away from each vertex. This implies that when ridges are present, the scaling ofL(Φ) at values of Φ much less than 1 but large enough to fully contain the vertices will be determined by the parts

12

of ridges which are closest to vertices. Ridges are also known to have a complicated spatial structure, but we assumethat the ridge solution converges to a simple scaling solution near the vertex, where the ridge length should becomeunimportant. It has been shown [22, 23] that the total bending and stretching energy of ridges in 2-sheets scale thesame way with manifold length scales and obey a virial theorem: the ratio of the total bending to stretching energy is1 to 5. The same virial ratio was also demonstrated for (m−1)-dimensional ridges in m-sheets [34]. We therefore inferthat to lowest order, the bending and stretching energy densities must follow identical scaling in the simple scalingregion near vertices. The virial relation, Eb = 5×Es, should also be evident in the ratio of scaling prefactors for thetwo energies. Furthermore, the total elastic energy in a ridge diverges as the length of the ridge becomes infinite [23],so the energy density along the ridge should not fall faster than Φ−1. Thus we expect the scaling behavior of ridgesto follow

Lb ∼ Φ−q

Ls ≈ 1/5Lb ∼ Φ−q

0 < q < 1. (14)

This dependence implies that strain energy has not condensed onto the vertices alone if ridges are present. Ingeneral, our assumption of near-isometry away from the defect set implies that strain will condense out of the bulk ofthe m-sheet. Thus we expect that the strain must condense onto the ridges and vertices – onto the full set K. Thismeans that, beginning at some Φc < 1 which marks the boundary of the ridge scaling region, there will be a morerapid drop-off (faster than Φ−1) of strain energy with volume away from the ridges.We can calculate the anticipated scaling exponent q above based on the anticipated scaling of the ridge width w(r)

at a distance r from a vertex, viz. w(r) = w(X)f(r/X), where X is the length of the ridge. Previous work [33] showsthat w(X) ∼ h(X/h)2/3 for m−1-dimensional ridges in m-sheets. We anticipate that w(r) ≪ w(X) when r ≪ X , andthat in this regime w(r) is independent of X . Then our scaling assumption implies w(r) ≈ h(r/h)2/3. The transversecurvature C(r) is as usual presumed to be of order 1/w(r). Lobkovsky [22] originally derived this scaling propertybased on more detailed assumptions. The curvature energy should be significant in a region of width w around thecenter of the ridge. The local energy density therefore scales as C2 ∼ r−4/3, while the high energy volume shouldgrow as Φ ∼ r × 1/C = r5/3. Thus the above curvature scaling leads to Φ−4/5 scaling for both Ls and Lb aroundthe vertex if it is the end-point of a ridge. This scaling was originally derived for 2-sheets embedded in 3 dimensions.Other work [33] suggests that the same scaling should hold for m-sheets in (m+1)-dimensional spaces. For m-sheets,ridges with spatial extent X in l long directions and width of the form w(x) given above in the remaining m − ldirections will occupy a total volume Xm(h/X)(m−l)/3. Compared with the total volume of the manifold, which isof order Xm, the high energy volume fraction is Φc ∼ (h/X)(m−l)/3 ≪ 1. Thus there is energy condensation ontoridges. For general dimensions, we reason that any balance of bending and stretching energies should lead to a virialrelation, and a virial relation in turn implies parallel scaling of the two energy densities. So, Eq. 14 should hold forall higher dimensional generalizations of ridges.

V. NUMERICAL METHODS

For the present study we have generalized the numerical approach of Seung and Nelson[2], modelling an m-sheet asan m-dimensional rectangular lattice and adding terms to the elastic energy to produce bending stiffness. Properlyspeaking, we simulate phantom m-sheets, which can pass through themselves. In the latter part of this section wediscuss the parameter range in which the phantom m-sheet behaves like a physical m-sheet, as well as the specialimplications of the phantom approximation on the structure of vertex singularities.Our manifold is a hypercubic array of nodes labeled by I ≡ i1, ..., im. Each node has a d-dimensional position

vector ~r(I). The relaxed lattice has a nearest-neighbor distance a. The lattice displacement from a site at I to anearby one can be expressed by a vector of m integers, ∆. It is convenient to define the lattice displacements ~u(I,∆),defined as the displacement between the node at site I and the one shifted by ∆

~u(I,∆) ≡ −~r(I) + ~r(I + ∆). (15)

The stretching energy U( ~R) for a 3-sheet (m = 3) is now defined as

U ≡ G∑

I

∑

∆=nn

(|u(I,∆)| − a)2

+cs∑

I

∑

∆=nnn

(|u(I,∆)| −√2a)2. (16)

Here the nn denotes the six nearest-neighbor sites ∆ = (±1, 0, 0), (0,±1, 0), and (0, 0,±1). The nnn sites are the12 second neighbor sites of the form (±1,±1, 0), etc. The weight coefficient cs assures that U is isotropic: i.e.

13

independent of the direction of strain relative to the lattice. We found by direct calculation of the elastic energy foruniform strain in the (1, 0, 0), (1, 1, 0) and (1, 1, 1) directions, minimized with respect to lateral expansion, that U [γ]was equal for the three directions of strain when cs = 1. The corresponding Poisson ratio is 1/4. By expanding Eq.(16) for small deviations of the 3-sheet from zero deformation and equating terms with those of Eq.(1), we infer thatfor our lattice µ = 4a2G and λ = µ.We use a discrete form of Eq. 2 to determine the curvatures in our simulated m-sheet. For each origin site I we

evaluate the diagonal elements ~κii from the nearest- neighbor separations:

~κii ≈1

a2[(~r(I + ∆i)− ~r(I))− (~r(I)− ~r(I−∆i))] (17)

The off-diagonal elements ~κij , i 6= j are computed in similar fashion from the next nearest-neighbor positions:

~κij ≈ 1

4a2[(

~r(I + ∆i +∆j)− ~r(I + ∆i −∆j))

−(

~r(I−∆i +∆j)− ~r(I−∆i −∆j))]

(18)

Once the curvature matrix ~κ is known for each site, we may compute the curvature energy Eb from Eq.(3). Tosave computational time we do not project the curvature vectors onto the normal space of the manifold at I. Thisamounts to including tangential components in the curvature tensor defined in Equation 2. It is the usual practice inlinear elasticity to neglect these terms because of their smallness [36], so leaving them in for computational efficiencydoes not introduce any significant change to the energy density profile.The sizes and elastic thicknesses of the lattices used in our simulations were arrived at through a trial-and-error

balancing of computational resources and data quality. We minimized elastic energy using an inverse gradient routine,which theoretically converges in ∼ N2 steps for a harmonic potential with N degrees of freedom [37]. However,experience shows that the convergence becomes much less efficient when we make the elastic sheets very thin, sincein this limit the total energy functional is highly non-linear and has large prefactors for the highest-order terms. Thiseffect in elastic simulations was described in [43], but their “reconditioning” approach to regaining fast convergencerequires too much computational overhead to be of use on large 3-dimensional lattices. The computational cost oflarger lattices must be balanced against the range of validity of the discrete lattice approximation. The lattice canonly accurately accommodate embeddings where the radius of curvature, 1/C, is locally much greater than the spacingbetween lattice points. We have no hope of maintaining accuracy at a vertex, which is a near singularity, but we tryto stay within an operating range where the sharpest features away from vertices have radii of curvature at least afew times the inter-lattice point spacing. This indirectly constrains the thickness of the elastic manifold we simulate,since features become sharper as the manifold is made thinner.Our standard simulational procedure was to begin with a lattice about 30 units on a side, since this was the smallest

lattice where fine features were clearly visible. After the elastic energy of the manifold was minimized on this lattice,we interpolated the result onto an 80 unit lattice and minimized again. Then we decreased the elastic thickness ofthe manifold on the larger lattice over a process of several minimizations. When the elastic thickness of the manifoldbecomes very small, the material becomes prone to falling into broad local minima with fine-scale crumpling whichconfuses the energy data. Slowly decreasing the thickness is a method to avoid this fine-scale crumpling. In most ofthe following sections we present the result of simulations on 80-unit lattices with an elastic thickness of ≈ 0.02 latticeunits. The entire process of generating each minimized lattice took up to several weeks on a 233 MHz, Pentium IIbased linux computer using a gcc compiler.We note that our lattice simulates a phantom m-sheet, which can pass through itself without penalty. Since the

energetic properties we study follow from local laws, and we stay in a thickness regime where curvature is non-singularalmost everywhere, the fact that our sheets are not self-avoiding does not affect the conclusions we draw from our data.The effect of the phantom m-sheet behavior on the dimensionality of vertex structures, where curvature does becomesingular, is discussed in Section X. In particular, as the thickness goes to zero, the minimum energy configurationsneed not converge to objects that have the local structure of a manifold. For example, in the vicinity of a vertex ina phantom sheet, as h → 0, the configuration might converge to a branched manifold (e.g., a cone which winds twicearound some axis).For geometries which generated several disclinations in a single manifold, we took special care with initial conditions

to insure that the system moved towards a symmetric final state. Relaxed states which contained a collection ofidentical vertices gave much cleaner scaling and always had a lower total energy than those which contained anensemble of vertices with different local structure. Thus, when we started the sheet in a state with several folds, weseparated the opposite sides of the folds slightly in globally symmetric ways to determine how they would relax.

14

FIG. 4: Equipotential surface of the spatially confining potential for a square 2-sheet embedded in 3 dimension.

(a) (b)

FIG. 5: Energy condensation map for spatially confined cubes. The cubes were X = 20 lattice sites wide and had elasticthickness h = .075X. Image (a) shows a surface of constant bending energy density in the material coordinate system for acube embedded in 4 dimensions. The surface encloses the ≈ 10% volume fraction with the highest energy concentration. Image(b) shows a surface of constant bending energy density for a cube embedded in 5 dimensions. This surface encloses the ≈ 7%volume fraction. The wireframes represent the edges of the cubes’ material coordinates.

(a) (b) (c)

FIG. 6: Method for creating disclinations. In (a), one edge of 2-sheet is folded and attached as shown. Points on the edgeare identified, but the curvature is not continued across the seam. Image (b) shows an equilibrium configuration of a 2-sheetconstructed as in (a) after elastic energy minimization (as in Fig 3). Illustration (c) shows how the same technique is used tomake a line disclination in an cubic 3-sheet.

15

VI. SPATIALLY CONFINED SHEETS

In this and the next three sections we report the results of our simulations. In this section we explore the distortionsresulting from spatially confining our 3-sheet in a contracting volume. Spatial confinement was simulated with an(R/b)10 potential, which acts essentially like a hard wall at a radius b. We tailored the potential to our cube-shaped3-sheet by making the equipotential lines nearly cubical in three spatial dimensions and spherically symmetric in allremaining dimensions. This reduced edge effects at the corners of our cubes. The exact form of the potential was

E ∝3

∑

i=1

(xi

b

)2

+

d∑

j=4

(xj

b

)2

5

. (19)

An equipotential surface of this potential for a 2-sheet in d = 3 is shown in Fig. 4.We began our simulations with the hard wall potential just outside the boundaries of the cube, then progressively

moved the walls inward on all sides until the geometrically confining volume had only half the spatial extent of theresting cube in any direction. The value of b was decreased in ten equal steps, with the lattice allowed to relax to anelastic energy minimum after each step. This procedure simulates a gentle confinement process, which allows appliedstress to propagate through the entire manifold volume instead of being caught in a strong ridge network at theoutside edges. Gentle confinement is essential to good convergence of the inverse gradient routines used to minimizethe elastic energy.Since the spatial confinement technique requires multiple inverse gradient minimizations for each simulation, it

is not computationally practical to run on large grids. Also, the data obtained from this method does not lenditself as well to numerical analysis, since the energy gradient from the tails of the hard wall potential mix with theelastic energy densities. Still, simulations performed on smaller grids (20 lattice units extent) show some interestingqualitative differences between confinement in d = 4 versus d = 5. As Fig. 5 shows, the regions of highest energydensity are well organized line-networks for d = 4 but are much more scattered and disorganized for d = 5. Ourarguments for the minimal dimensionality of D presented in Section IIIA predict a minimal dimension of 1 and 0respectively for 3-sheets embedded in 4 and 5 dimensions. If we assume that the high energy regions seen in Fig. 5surround parts of D, then the qualitative data supports these values for dimD.

VII. DISCLINATION PAIRS

To gain a better understanding of elastic energy condensation in m-sheets, we analyze several simpler forms ofdistortion. We first study pairs of disclinations. One way to create a disclination in a square 2-sheet is to join twoadjacent corners and the edge connecting them, as shown in Fig. 6(a). The disclination relaxes into a conical shapelike that shown in Fig. 6(b). Placing two or more conical disclinations in a 2-sheet in d = 3 causes ridges to form whichare apparently equivalent to those connecting vertex singularities in a confined sheet. A corresponding technique forcreating folds in a 3-sheet is to add line-like wedge disclinations into the manifold. We simulate line disclinations in3-cubes numerically by folding faces of an elastic cube down the center and connecting the two halves as shown inFig. 6(c).It can be shown by construction that the 3-cube in embedding space d > 3 can accommodate one line disclination

without stretching. One can construct such an embedding by bending each plane perpendicular to the line disclinationinto an identical cone. However, pure bending configurations for a 3-cube with two such line-disclinations will in generalrequire folds. It is energetically favorable for the cube to stretch to avoid singular curvatures, so we may expect thesheet to form ridges with the same degree of elastic energy condensation as in a physically confined 3-sheet.We begin our study of disclinations in general dimensions by simulating a 2-sheet with two sharp bends embedded

in either 3 or 4-dimensional space. From previous work [23] the 2-sheet in 3 dimensions should from a simple ridge– its expected behavior in 4 dimensions is not known. Next we turn our attention to 3-sheets, beginning with asimulation of a half-cube with a single line disclination embedded in either 4 or 5 dimensions. This simulation willverify the predicted scaling of a simple cone. Then, to induce elastic energy condensation we construct 3-cubes withtwo line disclinations at opposite cube faces and embed them in 4 or 5 dimensions. Since there is no guaranteethat our procedure will find the global energy minimum, we start the cubes in many different initial conditions. Weinvestigate the behaviors when the line disclinations are either parallel or perpendicular to one another in the materialcoordinates.

16

(a) (b)

(c) (d)

-4

-3

-2

-1

0

1

2

3

0 50 100 150 200

embe

ddin

g co

ordi

nate

(la

ttice

uni

ts)

material coordinate (lattice units)

FIG. 7: Equilibrium embedding coordinates for elastic 2-sheets with two 90o bends. The sheets were X = 200 lattice siteswide and had elastic thickness h = 2 × 10−4X. The bends were imposed by attaching opposite edges to a rigid right angleframe. Image (a) shows the three embedding coordinates for a 2-sheet in 3-dimensional space. Image (b) shows the same threeembedding coordinates for a 2-sheet in 4-dimensional space. Image (c) shows the fourth embedding coordinate (not shownin (b)) for the 2-sheet in 4-dimensions, plotted against the sheet’s material coordinates. In (c), the value of the embeddingcoordinate has been multiplied by 20 to enhance contrast. In (d) the embedding coordinate shown in (c) is plotted againstmaterial coordinate down the folding line for three different material thicknesses. The (+) symbols correspond to an elasticthickness of 1 lattice unit, the (×) symbols correspond to an elastic thickness of 0.1 lattice unit, and the () symbols correspondto an elastic thickness of 0.01 lattice unit.

A. 2-Sheet: Two Sharp Bends

The behavior of 2-sheets with two disclinations in 3-dimensional space has been studied extensively [22], and oursimulations of this geometry reproduced familiar results. However, we found for a variety of material thicknesses anddisclination geometries the behavior of the same 2-sheets embedded in 4 dimensions was remarkably different. Thedata presented here is for a sheet geometry which closely related to imposed disclinations and which displayed thesheet’s behavior particularly well. Instead of creating a disclination like that in Fig. 6 (a), we fold opposite edges ofthe sheet and attach them to rigid frames with sharp bends at their centers. Each frame keeps the edge straight witha 90o angle at its center point. The frames are free to translate or rotate in the embedding space. This boundarycondition is close to the conditions used to create “minimal” ridges in [22]. In that work Lobkovsky argued that theconfiguration of the sheet around a bending point on the edge will be much like that around a vertex. We found thatthe quantitative behavior of this boundary condition was consistent with that of imposed disclinations, but it allowedfor more flexibility. The equilibrium embeddings of sheets with this geometry are pictured in Fig. 7, Fig. 8 presentsplots of energy density versus area for 3 and 4 dimensional embeddings, and Fig. 9 plots local bending and stretchingenergy densities in the sheets’ coordinate systems.It is immediately evident from Figs. 9 that the stretching energy density in the region between the two sharp

17

(a)

1e-06

1e-05

1e-04

1e-03

1e-02

1e-01

0.001 0.01 0.1 1

ener

gy d

ensi

ty (

rela

tive

units

)

enclosed volume fraction

bendingstetching

(b)

1e-06

1e-05

1e-04

1e-03

1e-02

1e-01

0.001 0.01 0.1 1

ener

gy d

ensi

ty (

rela

tive

units

)

enclosed volume fraction

bendingstetching

FIG. 8: Energy density plots for elastic 2-sheets with two sharp bends. The sheets were X = 200 lattice sites wide and hadelastic thickness h = 2× 10−4X. In each graph the + symbols denote bending energy while the × symbols denote stretchingenergy. Energies are expressed in arbitrary units. Horizontal axes are area fraction Φ. Graph (a) shows local stretching energydensity Ls and bending energy density Lb versus area fraction Φ at or above this energy density from an embedding in 3dimensions. Graph (b) shows the same quantities from an embedding in 4 dimensions. In both graphs the straight lines arepower law fits to the bending and stretching energy densities. In all graphs the energy fits are to the region between 0.5% and2.0% volume fraction. In (a), the solid line is a fit to the bending energy density, with scaling exponent −0.61, and the dashedline is a fit to the stretching energy density, with scaling exponent −0.71. In (b), the solid line is a fit to the bending energydensity, with scaling exponent −0.66, and the dashed line is a fit to the stretching energy density, with scaling exponent −1.10.

bends is greatly diminished in the 4-dimensional embedding compared to the 3-dimensional embedding. For the latterembedding, the line of high stretching energy density in Fig. 9(b) marks the presence of the stretching ridge. However,there is no such stretching line in the 4-dimensional embedding energy map plotted in Fig. 9(d), even though Fig. 7(b)shows that there is still a folding line between the sharp bends in 4 dimensions. The energy plot in Fig. 8(a), for3-dimensional embedding, shows the parallel scaling of bending and stretching energies which is indicative of a ridge,but the energy plot in Fig. 8(b), for 4 dimensions, is more suggestive of cone-like scaling, since the stretching energyfalls twice as fast as bending energy away from the sharp bends.Examining the embedding coordinates of the manifold in 4 dimensions, we found that the sheet mainly occupied

only 3 on the 4 available spatial dimensions. Fig. 7(c) plots the value of the embedding coordinate with the lowestmoment of inertia (the moment for the entire manifold in this direction is four orders less than that in other directions,in a frame where the inertia tensor is diagonal). We believe this slight bubbling into the extra dimension acts as asink for compressive stress along the line connecting the sharp bends. Since this deviation is so small, one of the twonormals to the manifold lies mostly in this direction over the entire surface. The curvature shown in Fig. 7(c) is smallcompared to the major component of curvature across the folding line, and is nearly orthogonal to it, so it has littleeffect on the total bending energy. Yet, in the thin sheets we simulate, the resulting changes in the strain field affectthe stretching energy enormously.The bubbling discussed above is evidence of an interaction between the sharp bends, since such a configuration

is not seen for isolated disclinations or vertices and must be energetically less favorable than perfectly straight conegenerators. If the sharp bends interact in a way that depends on the distance between them relative to the elasticthickness, then there might be some analog of a higher dimensional ridge between them, with much weaker stretchingenergy. We did not see ridge-like parallel scaling in Fig. 8(b), but it is possible that the strain in this kind of ridge is soweak, and the virial ratio between bending and stretching is correspondingly so high, that the systems we simulatedwere dominated by a cone-like configuration near the sharp bends and not by the ridge between them. If this is thecase, our stretching energy graph shows only the initial energy fall-off away from the sharp bends and never reachesthe energy density value at which parallel scaling would commence. We can use the graph to put a lower limit onany possible virial relation by noting that cone-like scaling continues to at least 2% volume fraction, at which pointthe ratio between bending and stretch energy densities is ≈ 70. Thus, if the bending and stretching energies do scalewith elastic thickness, they should satisfy Eb > 70× Es.Following the derivation presented in Appendix A, we can use the virial relation to put limits on the scaling

exponents for the typical curvatures and strains on the ridge. For the above virial ratio, in the h → 0 limit the typicalmid-ridge curvature would increase more slowly than (X/h)1/35 and the typical ridge strain would fall faster than(h/X)34/35, where h is the elastic thickness and X is the length of the folding line. To test our scaling hypothesiswe probed the deviation into the normal direction shown in Fig. 7(c). We estimate that the inverse square of theheight of these bumps is proportional to the residual Gaussian curvature and therefore the strain along the ridge.

18

(a) (b)

(c) (d)

FIG. 9: Elastic energy density profiles in 2-sheet with two sharp bends pictured in Fig 7. The sheets were X = 200 latticesites wide. The elastic thickness of each was h = 2 × 10−4X. In each plot the height of the surface is proportional to energydensity in relative units and the x and y coordinates are the material coordinates in the manifold. The same energy units areused in all four plots. Plots (a) and (b) are for a 3-dimensional embedding, plots (c) and (d) are from a sheet embedded in 4dimensions. Plots (a) and (c) are the bending energies in the 2-sheets, plots (b) and (d) are the stretching energies.

In simulations of the same system at several different thicknesses, spanning two orders of magnitude, we could notdiscern a consistent change in the peak-to-peak height along this second normal (see Fig. 7 (d)). Since our ridgescaling arguments tell us we should see clear scaling of this peak-to-peak height with thickness, we conclude thateither we are not close enough to the thin limit for any potential scaling behavior to be evident, or the equilibriumconfiguration is really a higher dimensional variation of simple cone scaling, which is truly independent of elasticthickness and fold length. These tests were run at (h/X) ratios from 10−3 to 10−5, the entire range of thicknessesour simulations can handle and a region where 2-sheets in 3 dimensions show very clear thickness scaling. It is clearlybeyond our computational capabilities to resolve this potential scaling behavior.

B. 3-Sheet: Single Disclination

To verify our numerical predictions for the cone, we simulate an elastic half-cube with a single line disclination onone face. We use a 40× 80× 80 unit lattice. The minimum energy embedding is a virtually identical cone in all theplanes perpendicular to the line disclination. The radius of the cone ranges from 40 to

√2× 40 lattice units. Fig. 10

show the scaling of bending and stretching energy densities away from the disclination for both 4 and 5-dimensionalembeddings. In both cases the scaling exponents are very close to the theoretical values of −1 for bending and −2

19

(a)

1e-06

1e-05

1e-04

1e-03

1e-02

1e-01

0.001 0.01 0.1 1

ener

gy d

ensi

ty (

rela

tive

units

)

enclosed volume fraction

bendingstetching

(b)

1e-06

1e-05

1e-04

1e-03

1e-02

1e-01

0.001 0.01 0.1 1

ener

gy d

ensi

ty (

rela

tive

units

)

enclosed volume fraction

bendingstetching

FIG. 10: Energy density plots for elastic half-cubes with single line disclinations. The rectangular solids were X = 40 latticesites across perpendicular to the face with the disclination and 80 sites wide in the other directions. They had elastic thicknessh = 2.5× 10−4X. In each graph the + symbols denote bending energy while the × symbols denote stretching energy. Energiesare expressed in arbitrary units. Horizontal axes are volume fraction Φ. Graph (a) shows local stretching energy density Ls

and bending energy density Lb versus volume fraction Φ at or above this energy density from an embedding in 4 dimensions.Graph (b) shows the same quantities from an embedding in 5 dimensions. In both graphs the straight lines are power law fitsto the bending and stretching energy densities in the region between 2.0% and 10% volume fraction. In (a), the solid line isa fit to the bending energy density, with scaling exponent −0.95, and the dashed line is a fit to the stretching energy density,with scaling exponent −1.87. In (b), the solid line is a fit to the bending energy density, with scaling exponent −0.95, and thedashed line is a fit to the stretching energy density, with scaling exponent −1.87.

(a) (b)

FIG. 11: Energy condensation map for cubes with parallel disclinations. The cubes were X = 80 lattice sites wide and hadelastic thickness h = 2×10−4X. Image (a) shows a surface of constant bending energy density in the material coordinate systemfor a cube embedded in 4 dimensions. The surface encloses ≈ 10.0% volume fraction and shows energy condensation along theridge which spans the gap between the disclinations. Image (b) shows the same surface for a cube embedded in 5 dimensions.The wireframes represents the edges of the cubes’ material coordinates. Heavy lines mark the locations of disclinations.

for stretching for a cone with 2-dimensional symmetry. We are quite satisfied that the elastic lattice can accuratelyrepresent the cone around a single disclination.

C. 3-Sheet: Parallel Disclinations

Apart from boundary conditions, the cube with parallel disclinations has a natural symmetry along the direction ofthe disclinations. We found that for all initial conditions tested, energy minimization resulted in a final configurationwhich showed this same symmetry (see Fig. 11). The manifold has no strain or curvature in the direction parallelto the disclinations, and very similar configurations for all planes perpendicular to this direction. In principle, for

20

(a)

1e-05

1e-04

1e-03

1e-02

1e-01

1e+00

0.001 0.01 0.1 1

ener

gy d

ensi

ty (

rela

tive

units

)

enclosed volume fraction

bendingstetching

(b)

1e-06

1e-05

1e-04

1e-03

1e-02

1e-01

0.001 0.01 0.1 1

ener

gy d

ensi

ty (

rela

tive

units

)

enclosed volume fraction

bendingstetching

FIG. 12: Energy density plots for an elastic cube with parallel disclinations The cubes were X = 80 lattice sites wide and hadelastic thickness h = 2× 10−4X. In each graph the + symbols denote bending energy while the × symbols denote stretchingenergy. Energies are expressed in arbitrary units. Horizontal axes are volume fraction Φ. Graph (a) shows local stretchingenergy density Ls and bending energy density Lb versus volume fraction Φ at or above this energy density from an embeddingin 4 dimensions. Graph (b) shows the same quantities from an embedding in 5 dimensions. In both graphs the straight linesare power law fits to the bending and stretching energy densities. In graph (a) the bending energy fit is to the region between3.0% and 8.0% volume fraction and the stretching energy fit is to the region between 2.0% and 5.0% volume fraction. In graph(b) the fits are to the region between 1.0% and 4.0% volume fraction. In (a), the solid line is a fit to the bending energy density,with scaling exponent −0.63, and the dashed line is a fit to the stretching energy density, with scaling exponent −0.48. In (b),the solid line is a fit to the bending energy density, with scaling exponent −0.93, and the dashed line is a fit to the stretchingenergy density, with scaling exponent −1.59.

embedding in d dimensions, the configuration in each of the perpendicular planes is identical to that which we wouldexpect for an elastic 2-sheet with the same thickness to length ratio embedded in d − 1 dimensions. In practice, wefind that the extra material dimension adds an additional stiffness against fine scale crumpling which often confusessimilar simulations in 2-sheets.For 4-dimensional embedding the equilibrium configuration is a “stack of ridges,” which shows the same energy

scaling as a ridge in 3 dimensions. Fig. 12 (a) is a plot of energy density versus volume for a 3-cube with paralleldisclinations in 4 dimensions. Within the high energy region encompassing ≈ 2 − 8% volume fraction, the ratio ofbending energy density to stretching energy density at a given volume fraction is ≈ 6.2. This number is consistentwith the theoretical energy ratio of 5. In this volume range the plots also confirm the lack of condensation along theridge of both bending and stretching energies as well as the identical scaling of these energies. For the entire region upto ≈ 30% volume fraction the bending and stretching energy have roughly the same dependence on volume, thoughthey don’t fit a clean scaling exponent for any extended region. The sharp drop-off of the dominant energy above30% volume fraction shows the significant condensation of energy around the ridge structure.In contrast to the ridge-scaling in 4 dimensions, the energy scaling behavior of the sheet embedded in 5 dimensions

appears cone-like. The stretching scaling exponent of −1.59 indicates that the stretching energy is condensed at thevertices, while the bending exponent of −0.93 is consistent with −1, the predicted bending scaling of isolated linedisclinations. The scaling data, as well as the lack of a strong ridge region in Fig 11(b), indicate that the scalingaround each disclination is not strongly influenced by interaction between the two disclinations. The scaling resemblesthat around the isolated disclination reported in Section VIIB.

D. 3-Sheet: Perpendicular Disclinations

A typical spatial energy distribution after energy minimization for a cube with perpendicular disclinations embeddedin d = 4 is shown in Fig. 13(a). A common feature of all the cubes embedded in 4 dimensions is the spontaneousappearance of additional line-like vertex structures. Spanning the volume between vertices and disclinations are 2-dimensional ridges. This result is consistent with the d = 4 geometrical confinement simulations detailed in [34].Fig. 14(a) shows the decay of local energy density with volume away from ridges and vertices for an elastic cube withnon-parallel disclinations in 4 dimensions. The highest-energy regions correspond to the imposed disclinations andspontaneous vertex network. Lower energy regions correspond to ridges. The region between 1% and ≈ 5% volumefraction shows smooth scaling of bending and stretching energy densities with volume in a region dominated by thehigh energy part of ridge structures (where they join at vertices). In this region the bending energy density scales

21

(a) (b)

FIG. 13: Energy condensation map for cubes with perpendicular disclinations. The cubes were X = 60 lattice sites across, 90lattice units wide in the directions parallel to the disclinations, and had elastic thicknesses h = 6×10−4X (a) and h = 3×10−4X(b). Image (a) shows a surface of constant bending energy density in the material coordinate system for a cube embeddedin 4 dimensions. Image (b) shows a surface of constant bending energy density for a cube embedded in 5 dimensions. Theenergy density value encloses ≈ 3% volume fraction in image (a) and shows the spontaneous vertex lines which arise betweendisclinations. In image (b) the equal energy density surface encloses ≈ 10% volume fraction and shows the growth of the highenergy region around the disclinations without additional vertices evident between them. The wireframes represent the edgesof the cubes’ material coordinates. Heavy lines mark the locations of disclinations.

(a)

1e-04

1e-03

1e-02

1e-01

1e+00

0.001 0.01 0.1 1

ener

gy d

ensi

ty (

rela

tive

units

)

enclosed volume fraction

bendingstetching

(b)

1e-06

1e-05

1e-04

1e-03

1e-02

1e-01

1e+00

0.001 0.01 0.1 1

ener

gy d

ensi

ty (

rela

tive

units

)

enclosed volume fraction

bendingstetching

FIG. 14: Energy density plots for the elastic cubes with non-parallel disclinations in Fig 13. The cubes were X = 60 latticesites across, 90 lattice units wide in the directions parallel to the disclinations, and had elastic thicknesses h = 6× 10−4X (a)and h = 3 × 10−4X (b). In each graph the + symbols denote bending energy while the × symbols denote stretching energy.Energies are expressed in arbitrary units. Horizontal axes are volume fraction Φ. Graph (a) shows local stretching energydensity Ls and bending energy density Lb versus volume fraction Φ at or above this energy density from an embedding in 4dimensions. Graph (b) shows the same quantities from an embedding in 5 dimensions. In (a), the solid line is a power lawfit to the bending energy density in the region between 1.0% and 5.0% volume fraction, with scaling exponent −0.77, and thedashed line is a power law fit to the stretching energy density in the region between 3.0% and 10% volume fraction, with scalingexponent −0.91. In (b), the solid line is a power law fit to the bending energy density in the region between 1.0% and 10.0%volume fraction, with scaling exponent −0.90, and the dashed line is a power law fit to the stretching energy density in theregion between 1.0% and 5% volume fraction, with scaling exponent −1.65.

with exponent −0.77 and the stretching with exponent −0.91. The values of the scaling exponents in this region arereasonably close to each other and to the theoretical scaling of −4/5 derived in Section IV. The scaling is clearlydistinct from that of a cone, where the stretching energy density is expected to fall faster than Φ−1.The behavior of the same cubes placed in d = 5 was quite different (see Fig. 13 (b)). For this embedding there

is no spontaneous ridge-vertex network between the imposed disclinations. The difference in structure is reflected inthe energy density plot, Fig. 14(b). In the volume fraction which typically encompasses high energy structures apartfrom vertices and disclinations, the bending energy scales with volume with an exponent of −0.90 while the stretching

22

(a) (b)

FIG. 15: Two views of a 3-dimensional projection of the embedding coordinates for a 2-slice from the material coordinates ofa cube with non-parallel disclinations embedded in 5 dimensions. The material coordinate slice contains one disclination alongthe edge marked with the heavy line in (a). It is perpendicular to the other disclination, which is marked with a heavy line in(b). The views were chosen to show the cone-like geometry about the disclination which the plane intersects at a point.

energy scales with an exponent of −1.65. These numbers indicate the dominance of conical scaling near the vertices.They are similar to the scaling exponents for the cube with parallel disclinations in 5 dimensions. Fig.15 shows thatthe embedding appears to be locally conical around a disclination, without any evidence of folds. We surmise thatthe sheet has relaxed to a configuration where the cones around each line vertex interpenetrate without interactingstrongly. This result demonstrates that in higher dimensional crumpling, multiple vertices can exist in a sheet withoutrequiring folds, for some geometries.For this simulated geometry and several of the following, we present data for thicker sheets in 4 dimensions than in

5. This is because the ridges found in 4 dimensions become very sharp as h gets smaller, and we typically present thethinnest data which does not show signs of finite lattice effects (like those seen below in Fig 17(a)). For 5-dimensionalembeddings, decreasing the thickness typically shifts all the stretching energy densities upward and thereby extendsthe volume of cone scaling visible before the stretching energy density fades to background levels. We thus chose topresent data from thinner sheets for 5 dimensions. We found that ridge scaling becomes more distinct as the sheetbecomes thinner, so the use of thinner sheets in 5 dimensions only strengthens our claim that there is no evident ridgestructure in 5 dimensions.The analog of ridges in the 3-torus in d = 4 are again planes of high elastic energy. These ridge structures meet

in vertex-lines of very high elastic energy. Fig 17(a) shows the decay of local energy density with volume away fromridges and vertices for a 3-torus in 4 dimensions. We fit a power law to the region which we identify as parts of ridgesnear the vertex structures. In this region the energy densities scale with an exponent of −0.87 for bending energyand −0.88 for stretching. These values are consistent with the theoretical ridge scaling exponent of −4/5 derived inSection IV. Within the ridge scaling volume the ratio of bending energy density to stretching energy density is 8.5,within a factor of two of the known value, 5, for 2-sheets in 3-dimensional crumpling.

VIII. TOROIDAL CONNECTIVITY

Toroidal connectivity was simulated numerically by defining the lattice displacement vector, ∆, such that oppositefaces of our cubic array had nearest-neighbor connections. The resulting connectivity was everywhere isotropic, withno borders or disclinations. Simulations were run for 3-tori embedded in 4,5 and 6 spatial dimensions. Initial conditionswere either random or chosen to be very symmetric or close to possible energy minima. In all cases when d = 4 or5, minimization of the elastic energy resulted in energy condensation and the formation of high-energy networks. For6-dimensional embeddings the elastic energy was many times smaller than in lower dimensions and was uniformlydistributed over the manifold. Figure. 16 compares the energy condensation networks for 3-tori embedded in either 4or 5 dimensions. Although the network is more extensive for d = 4, in either dimension high-energy structures havecomparable energy densities. We simulated this geometry beginning from many different initial conditions and theresulting final configurations showed varying degrees of asymmetry between the vertices. The configuration presentedin Fig. 16(a) showed the highest degree of symmetry of all our 4-dimensional torus simulations – its total elastic

23

(a) (b)

FIG. 16: Energy condensation maps for 3-tori. The tori were made from cubes of width X = 80 lattice sites and with elasticthicknesses h = 1 × 10−3X (a) and h = 2.5 × 10−4X (b). Image (a) shows a surface of constant bending energy density inthe material coordinate system for a 3-torus embedded in 4 dimensions. In this simulation the initial conditions were chosento favor a symmetric relaxed state. Image (b) shows a surface of constant bending energy density for a 3-torus embedded in 5dimensions. The surfaces encloses ≈ 2.5% volume fraction in (a) and ≈ 2.8% volume fraction in (b). The wireframes representthe edges of the cubes’ material coordinates.

(a)

1e-04

1e-03

1e-02

1e-01

1e+00

0.001 0.01 0.1 1

ener

gy d

ensi

ty (

rela

tive

units

)

enclosed volume fraction

bendingstetching

(b)

1e-04

1e-03

1e-02

1e-01

1e+00

0.001 0.01 0.1 1

ener

gy d

ensi

ty (

rela

tive

units

)

enclosed volume fraction

bendingstetching

FIG. 17: Energy density plots for the 3-tori in Fig 16. The tori were made from cubes of width X = 80 lattice sites and withelastic thicknesses h = 1 × 10−3X (a) and h = 2.5 × 10−4X (b). In each graph the + symbols denote bending energy whilethe × symbols denote stretching energy. Energies are expressed in arbitrary units. Horizontal axes are volume fraction Φ.Graph (a) shows local stretching energy density Ls and bending energy density Lb versus volume fraction Φ at or above thisenergy density from an embedding in 4 dimensions. The several plateaus in the high energy part of the plot are an artifact ofthe discrete lattice. They reflect the nearly identical geometry of the points on and adjacent to the vertices, which make up ameasureable fraction of the manifold volume. Graph (b) shows the same quantities as in (a) for an embedding in 5 dimensions.In (a), the solid line is a power law fit to the bending energy density in the region between 2.0% and 10% volume fraction, withscaling exponent −0.87, and the dashed line is a power law fit to the stretching energy density in the region between 2.0% and6.0% volume fraction, with scaling exponent −0.88. In (b), the solid line is a power law fit to the bending energy density inthe region between 0.5% and 10.0% volume fraction, with scaling exponent −0.56, and the dashed line is a power law fit to thestretching energy density in the region between 0.5% and 10% volume fraction, with scaling exponent −1.02.

energy was ∼ 25% less than that of configurations which broke symmetry, so we believe it is most likely the trueground state of the system. Our energetic analysis was performed on this configuration.The energy structures of tori embedded in d = 5 were qualitatively different from those embedded in d = 4 (refer

again to Fig. 17). For d = 5, the structures corresponding to vertices appear point-like instead of line-like. Themajority of the total energy density is concentrated around these point-like vertices. Between vertices we were ableto see smaller, line-like energy concentrations of elastic energy which could correspond to ridge structures. However,these regions occupied a miniscule volume in the manifold. The predominant energy structures were more symmetricand were centered around vertices.

24

(a) (b)

(c) (d)

FIG. 18: Energy condensation maps for 3-cubes with center points of opposite faces attached. The cubes were X = 80 latticesites wide and had elastic thickness h = 2.5× 10−4X. Images (a) and (b) show surfaces of constant bending energy density inthe material coordinate system for a 3-cube embedded in 4 dimensions. In this simulation the initial conditions were chosento favor a symmetric relaxed state – many stable configurations show pronounced symmetry breaking in one direction. Images(c) and (d) show surfaces of constant bending energy density for a 3-cube embedded in 5 dimensions. The surfaces in (a) and(c) encloses 0.1% volume fraction while those in (b) and (d) encloses 1%. The wireframes represent the edges of the cubes’material coordinates.

Energy density versus volume is plotted in Fig 17(b) for an 80 lattice unit 3-torus embedded in d = 5. Smoothenergy scaling begins at about 0.5% volume fraction and holds for up to ≈ 10% of the total volume. Within thesehigh energy regions the bending energy density scales with volume with an exponent of −0.56, whereas stretchingenergy density dies off more quickly, with a scaling exponent of −1.02. This is consistent with our simulations ofline disclinations in 5 dimensions, since the stretching energy falls off nearly twice as fast as the bending energy.The number of vertices present in the manifold would lead us to expect ridges, but as we saw in Sec.VIIC, ridgescannot be resolved by energy scaling alone in 5 dimensions at elastic thicknesses accessible to our simulations. Thescaling exponents above are closer to the expected exponents of −2/3 and −4/3 for conical scaling around a point-likedisclination than to any other kind of scaling behavior we know. The embedding is probably close to this form ofcone near the vertices.Figure 18 shows the pattern of energy condensation in 3-cubes with attached opposite faces after energy minimiza-

tion. For 4-dimensional embeddings the surfaces of constant energy density enclose line-like regions which traversethe cube as seen in Figure 18 (a) and (b). The high energy regions appear line-like all the way up to the highest valuesof the energy density. In contrast, the surfaces of constant energy density for 5-dimensional embeddings form a seriesof shells, as seen in Figure 18 (c) and (d), whose typical diameters increase with decreasing energy density value. Theenergy density at the surface in Figure 18 (c), which encloses 0.1% volume fraction, is nearly an order of magnitudegreater than the energy density at the surface of Figure 18 (d), which encloses 1% volume fraction and just touches

25

(a)

1e-04

1e-03

1e-02

1e-01

1e+00

0.001 0.01 0.1 1

ener

gy d

ensi

ty (

rela

tive

units

)

enclosed volume fraction

bendingstetching

(b)

1e-04

1e-03

1e-02

1e-01

1e+00

0.001 0.01 0.1 1

ener

gy d

ensi

ty (

rela

tive

units

)

enclosed volume fraction

bendingstetching

FIG. 19: Energy density plot for the elastic cubes with center points of opposite faces attached pictured in Fig 18. The cubeswere X = 80 lattice sites wide and had elastic thickness h = .001X. In each graph the + symbols denote bending energy whilethe × symbols denote stretching energy. Energies are expressed in arbitrary units. Horizontal axes are volume fraction Φ.Graph (a) shows local stretching energy density Ls and bending energy density Lb versus volume fraction Φ at or above thisenergy density from an embedding in 4 dimensions. Graph (b) shows the same quantities from an embedding in 5 dimensionsIn (a), the solid line is a power law fit to the bending energy density in the region between 3.0% and 10% volume fraction, withscaling exponent −1.02, and the dashed line is a power law fit to the stretching energy density in the region between 2.0% and9.0% volume fraction, with scaling exponent −0.75. In (b), the solid line is a power law fit to the bending energy density inthe region between 0.2% and 2.0% volume fraction, with scaling exponent −0.65, and the dashed line is a power law fit to thestretching energy density in the region between 0.2% and 2.0% volume fraction, with scaling exponent −1.26.

the outside edges of the cube. These data clearly support our assertion that point-like vertex structures are possible in5-dimensional embeddings. At the same time, they are consistent with conjectures [22] that the high energy regions in4-dimensional embeddings are line-like. It may be noted that the 5-dimensional embedding is asymmetric, while the4-dimensional embedding has a high degree of symmetry. We found that the elastic manifolds always spontaneouslybroke symmetry in 5 dimensions, but the minimum energy configuration we could find in 4 dimensions was perfectlysymmetrical.

IX. SINGLE FOLD (BOW CONFIGURATION)

In our final set of simulations we set our boundary and initial conditions to create single, point-like vertices. Ouraim was to verify scaling predictions for vertex deformations in 4 and 5 dimensions and to show clearly the existenceof point-like vertex structures in 5-dimensional embeddings. According to the relations presented in Section II, avertex is expected to have high Gaussian curvature. In more specific terms this means a vertex is a locus of strongcurvature in at least two material directions along the same normal vector. Therefore, to force the existence of exactlyone vertex in 5 dimensions (where every manifold point has two independent normals) we searched for a minimalset of boundary conditions which necessitated that some points in the manifold have curvature in all three materialdirections. Our most successful simulations, presented here for 4 and 5-dimensional embeddings of a 3-cube, had onlythe center points of opposite faces attached. This geometry caused vertices to form while leaving the majority of theboundary free.Plots of energy density versus enclosed volume fraction for this geometry are presented in Figure 19. For 4-

dimensional embeddings the bending energy density does not scale with a simple exponent in the volume range between1% and 10% volume fraction which we associate with the high energy region of ridges. To get a representative valueof the energy drop-off we fit a power law to the region between 3% and 10% volume fraction and find an exponent of−1.02. The stretching energy scales more cleanly in the region between 2% and 9% volume fraction, with an exponentof −0.75. The stretching energy exponent is close to the expected value of −4/5 and is smaller in magnitude than −1,so we can safely identify the scaling as ridge-like. When viewing the equal energy surfaces at larger volume fractions(not shown here) we saw a good deal of secondary structure in the ridges themselves, which could explain the manyfeatures in the energy density dependence.For the 5-dimensional embedding we fit a simple power law to the region of the graph which enclosed less than

≈ 2% volume fraction, since above this enclosed volume fraction the surfaces shown in Figure 18 (c) and (d) intersectthe boundary of the cube. [47] In this region the bending energy density scales with an exponent of −0.65 whilethe stretching energy density scales with an exponent of −1.26. These numbers are consistent with the theoretical

26

geometry m d vertex dimension scaling spontaneous assymetry

two sharp bends 2 3 n/a ridge n/a

2 4 n/a cone n/a

isolated disclinations 3 4 n/a cone n/a

3 5 n/a cone n/a

‖ disclinations 3 4 n/a ridge n/a

3 5 n/a cone n/a

⊥ disclinations 3 4 1 ridge n/a

3 5 n/a cone n/a

torus 3 4 1 ridge no

3 5 0 cone yes

bow 3 4 1 ridge no

3 5 0 cone yes

TABLE I: Summary of simulational results. The first three columns list the geometry of the simulated sheet and the spatialand embedding dimensions. The next column list the dimension of any spontaneous vertex structures. The next column tellswhether the observed scaling was cone-like or ridge-like. The final column tells, for the cases where the boundary conditionsdid not explicitly break the 3-dimensional symmetry of the cube, whether the energy minimum was a symmetric state in themanifold coordinates. The six manifold geometries, in the order presented here, are: two point disclinations in a square sheet,a single line-disclination at one cube face, parallel line disclinations on opposite cube faces, perpendicular line disclinations onopposite cube faces, toroidal connectivity of a 3-dimensional manifold, and the attachments of the center points of oppositecube faces to each other.

5-dimensional cone scaling exponent of −2/3 and −4/3 derived in Section IV.

X. DISCUSSION

In the numerical simulations reported in this paper we have investigated the behavior of 2 and 3-dimensionalmanifolds embedded in dimensions 3 and greater, subject to a variety of boundary conditions which cause crumpling.The results of our simulations, which are summarized in Table I, show a consistent dependence of the crumplingresponse on d−m, the difference between the dimensionality of the embedding space and the that of the sheet. Thebehavior summarized in Table I can be described by the following general principles:

1. For all the boundary conditions we consider, the dimensionality of spontaneous verticesD is 2d−m−1, suggestingthat the vertex dimensionality is always given by the lower bound from the arguments in Section IIIA thatyield dim(D) ≥ 2m− d− 1.

2. If d = m + 1, the details of the curvature defect set K determine the nature of the energy condensation. ForD = K (no folding lines) we have the cone scaling discussed in Section IV, and for K − D 6= ∅ (folding linespresent) we find ridge scaling.

3. If d ≥ m+ 2 we always find cone scaling if K 6= ∅.

We believe that these principles are true in general. Consequently, they give explicit predictions for the behaviorof elastic sheets in higher dimensions.

A. Effect of Embedding Dimension on Defect Dimension

Our data consistently supports the arguments presented in Section IIIA that for an m-dimensional manifold ind-dimensional space with d < 2m, the dimensionality of the set D of vertex singularities will be greater than or equalto 2m − d − 1. In fact, we found that all spontaneous vertex structures had dimensionality 2m − d − 1 identically.For 5-dimensional embeddings of 3-sheets in which we did not explicitly make line-like disclinations, the manifoldswere able to relax to configurations wherein D was small and point-like. In contrast, when the embedding space was4-dimensional for the same manifolds and boundary conditions, D was always line-like, terminating only at materialboundaries. It is worth noting that in all cases where D was 1-dimensional, it was also piecewise flat – it is easy to

27

argue that in order to minimize its spatial extent, D will in general be flat. Though our dimensionality argumentsapply to only asymptotically thin sheets, the predictions appear well obeyed even for sheets as thick as 1/30 of theirwidth. The dimensional behavior is thus much more robust than the ridge scaling behavior.Because of the phantom nature of our simulated sheet, we found that favorable energy configurations often had local

geometries at vertices in which the sheet passed through itself – the most common example was branched manifoldsat vertices. The arguments for scaling and the minimum dimensionality of D are not affected by whether or not thesheets are phantom. However, the set of boundary conditions that produce non-phantom sheets with minimal vertexdimensionality may be more limited than for phantom m-sheets.

B. Effect of Embedding Dimension on Scaling

In Section IV we made predictions for the energetic scaling outside of vertex structures for a sheet whose thicknesswas much smaller than its width. However, we did not analytically address how thin the sheet must be before itdisplays “thin limit” behavior. Our observations of the typical progress of a numerical simulation lend some insightinto the approach to this asymptotic limit. In the process of relaxing our sheets, we often vary the thickness h, whichshows us how the energy distribution depends on h. We were not able to vary h enough to directly observe any scalingbehavior with h. Even our smallest h’s show little enough of the desired asymptotic behavior, and increasing h onlyblurs this behavior beyond recognition. However, the qualitative behavior with h is consistent with our conclusions.The reported results are for the smallest h values we could reliably attain. When h is made larger, the main effect isto reduce the dynamic range in our energy density plots. It makes the energy spread more uniformly over the sheet.Where ridge scaling is observed, increasing h reduces the proportion of stretching energy, as we’ve previously observedin 2-sheets[22]. Finally, increasing h reduces the observed effect of embedding dimension. The clear differences betweenthe stretching and bending energy profiles in 5 dimensions become less distinct as h is increased. This is as expected.Embedding dimensions should have less effect if a sheet is thicker. These behaviors add to our confidence that thescaling behavior we report becomes more, not less, distinct as we reduce h.Our derivation of cone scaling was based on very simple and well-founded assumptions, so it was not a surprise

that cone scaling was so clearly visible in geometries where we expected to find it. In all simulations where there wasonly one disclination structure in a 3-dimensional manifold, the observed scaling was consistent with our predictionsbased on cones. A single line disclination in either 4 or 5 dimensions produced a cone-structure with scaling like thatof a simple cone in a 2-sheet. In the simulations presented in Section IX, embeddings in 5 dimensions produced apoint-like vertex and energy scaling close to our predictions for double-cone scaling (where there is curvature on thesame order in both material directions perpendicular to the cone generators). The success of these predictions assuresus that the cone-structure is well understood and that, for the case of line disclinations, the description of a relaxedconfiguration in terms of a stack of 2-sheet embeddings can be accurate.Our simulations verified the formation of ridges in 2-sheets in 3 dimensions and in 3-sheets in 4 dimensions, as

witnessed in other studies [3, 34]. As expected, planar ridge structures spanning the gaps between linear vertexstructures in d = 4 had the same energy-bearing properties as their 3-dimensional equivalents. However, for 2-sheetsin 4 dimensions and 3-sheets in 5 dimensions we consistently found no ridge energy structures, even along whatappeared to be folds. In Fig. 7(c) it is apparent that the elastic sheet deflects slightly into the fourth dimension inorder to relieve the strain along the ridge-center. The way this ridge decreases the strain along its midline showsthe essential difference between embeddings of m sheets in m + 1 dimensions and in all greater dimensions. In thegeometric von Karman equations, Eq. 9, the sources of strain are sums of curvatures along different normals. If thereis more than one normal direction, then there is the possibility of cancellation between curvature terms for any given2-dimensional hyperplane. On the 4-dimensional fold the manifold bubbles into the extra dimension, creating positiveGaussian curvature which counters the negative Gaussian curvature of the saddle-shaped peak-to-peak ridge profile(see for example Fig. 3). We do not assume that the cancellation of lowest order terms is perfect, but we showed inSection VIIA that the stretching energy is diminished so greatly relative to the bending energy that ridge-like scalingis completely masked by conical scaling near vertices (It is notable that, based on the scaling arguments derived inAppendix A, in the limit of very large virial ratios the dominant curvature terms become increasingly insensitive tothe elastic thickness – much like they are in cones). We hope to return to this subject in future research, with the aimto find analytic expressions for the embedding of a 2-sheet with two disclinations in 4 spatial dimensions and derivethe thin limit scaling directly.

28

C. New Questions Raised by This Work

We believe that the detailed structure of the novel point-like vertices in 3-sheets deserves further study. For example,in Section IX we touched on the fact that these point-vertices are loci of folding in all three material directions. Sincethe three material directions share only two normals and are constrained by the boundary conditions that the sheetbe isometric outside the vertex point, it is likely that the angles each direction folds by can only take discrete sets ofvalues – not unlike the discretization of disclination angles in a lattice, but driven purely by spatial geometry for anotherwise continuous medium. This discetization would probably be more pronounced in sheets that are not phantom,since branched manifolds at vertices add more degrees of freedom.Much could also be learned by viewing these new crumpling phenomena as a mathematical boundary-layer problem,

since they display several new and intriguing features in this light. Our problem belongs to the class of variationalproblems given by a singularly perturbed energy functional Eh with small parameter h. One approach to analyzingsuch variational problems is by identifying the boundary layers and then determining the appropriate solutions throughmatched asymptotics. Example of this “local” approach, as applied to elastic 2-sheets are the analysis of ridges inRef. [22] and of the vertices in [10].A contrasting “global” approach to these problems is through the notion of Γ – convergence [44, 45]. This approach

calls for the identification of an appropriate asymptotic energy E∗ that gives the energy of a configuration in theh → 0 limit. This asymptotic energy functional is called the Γ limit of the functional Eh as h → 0. If the Γ limitexists, the configuration of the minimizers for a small but nonzero h is then deduced by finding the minimizers for E∗

and observing that the minimizer for nonzero h is close to the minimizer for E∗. Note that this approach is similar inphilosophy to our arguments in Sec. III where we deduced the structure of the vertex regions for h > 0, by consideringisometric immersions which are relevant for h = 0.Since these singularly perturbed variational problems show energy condensation, in the limit h → 0 all the energy

concentrates on to a defect set, which we denote by B. Consequently, the appropriate asymptotic energy should alsobe defined for singular configurations, and it should depend on the defect set B, and the configuration u outside B.Thus, the asymptotic energy is given by a functional E∗ = E∗[h, u,B]. Examples of this type of analysis are theanalysis of phase separation in Refs. [18, 19] and the asymptotic folding energy in the context of the blistering of thinfilms [25, 26, 29]. In both these cases, the asymptotic energy scales with the small parameter h as E∗ ∼ hα for afixed α. Also, the Γ limit is local, in that the asymptotic energy is given by integrating a local energy density overthe defect set. This is in sharp contrast to the behavior of elastic manifolds. For elastic manifolds, the asymptoticenergy depends on two kinds of defect sets, the strain defect set D and the curvature defect set K. The exponent αthat gives the scaling of E∗ with h is not fixed, but depends on whether or not D = K. Finally, in the case of ridgescaling, we have the following two interesting features:

1. The width of the boundary layer around the ridge depends on both the small parameter h and the length of theridge X .

2. The energy of the ridge scales as h5/3X1/3 and is not linear in the size of the ridge. Therefore, the asymptoticenergy of the ridge is not given by an integral of a local energy density over K.

These features imply that the Γ limit E∗, if it exists, is non-local. It is therefore very interesting to carry out a rigorousanalysis of the Γ limit for the elastic elastic energy functional in Eq. 6. This analysis will probably involve new ideasand techniques.

Conclusion

In this paper we have found two important results for crumpled sheets. First, we have shown that if the spatialdimension d is greater than m+1, the stretching elastic energy condenses onto vertex structures, while for the specialcase d = m + 1 it condenses onto ridges as well. Second, we have provided evidence that when d < 2m the straindefect set in a crumpled sheet has dimension at least 2m − d − 1.For higher-dimensional manifolds with m > 3 onemay imagine further forms of energy condensation as the embedding dimension increases. Such manifolds could revealfurther surprises, as the present study did. Like gauge fields, elastic manifolds have revealed distinctive ways in whichsingularities in a field may interact at long range. To explore further the conditions and forms of this interactionseems worthwhile.

29

Acknowledgments

The authors would like to thank Bob Geroch, L. Mahadevan, Bob Kohn and Felix Otto for helpful and enlighteningconversations. This work was supported in part by the National Science Foundation under Award number DMR-9975533 Account 5-27810.

APPENDIX A: DERIVATION OF VIRIAL RELATION

Here we derive the relation between the energy scaling exponents and the virial ratio of bending to stretchingenergies on an elastic ridge. This derivation is a generalization of the derivation presented in [23].We assume for simplicity that on a ridge, the elastic bending energy is dominated by the contribution of the main

curvature accross the ridge, and this curvature is approximately constant for the entire length of the ridge with atypical value C. For a simple ridge of length X in a two dimensional manifold, the ridge curvature is significant in aband of width w = 1/C transverse to the ridge, so the total bending energy of the ridge is approximately

Eb ≈ µh2C2wX = µh2CX, (A1)

where µ is the elastic modulus of the material and h is the thickness.If we assume that the ridge has a single dominant component of strain which also extends for a typical width w

and has typical value γ, then the total stretching energy of the ridge is

Es ≈ µγ2wX = µγ2X/C. (A2)

We can write the total elastic energy along the ridge as

E ≈ µX[

h2C + γ2C−1]

. (A3)

Now, if we make a scaling ansatz, γ ∼ (XC)−α, the energy becomes

E ∼ µX[

h2C +X−2αC−(2α+1)]

. (A4)

We may now find the minimum energy balance by setting the derivative dE/dC equal to zero:

h2 − (2α+ 1)X−2αC−(2α+2) = 0 (A5)

⇒ C ∼ X−1 (h/X)−1/(α+1)

(A6)

⇒ γ ∼ (h/X)α/(α+1) (A7)

From the first of the above equations, it is clear that the virial ratio is related to α by Eb = (2α+ 1)Es.

[1] D. R. Nelson, and L. Peliti, “Fluctuations in Membranes with Crystalline and Hexatic Order” J Phys (Paris) 48, 1085(1987).

[2] H. S. Seung and D. R. Nelson, “Defects in Flexible Membranes with Crystalline Order” Phys Rev A 38, 1005 (1988).[3] A. E. Lobkovsky, T. A. Witten, “Properties of Ridges in Elastic Membranes” Phys Rev E 55, 1577 (1997).[4] M BenAmar, Y Pomeau, “Crumpled Paper” P Roy Soc Lond A Mat 453, 729 (1997).[5] M BenAmar, Y Pomeau, “Buckling of Thin Plates in the Weakly and Strongly Nonlinear Regimes” Phil Mag B 78, 235

(1998).[6] E. Cerda, S. Chaieb, F. Melo, L. Mahadevan, “Conical Dislocations in Crumpling” Nature 401, 46 (1999).[7] A. Boudaoud, P. Patricio, Y. Couder, M. Ben Amar, “Dynamics of singularities in a constrained elastic plate” Nature 407,

718 (2000).[8] L. A. Cuccia, R. B. Lennox, “Molecular Modeling of Fullerenes with Modular Origami” Abstr Pap Am Chem S 208, 50

(1994).[9] S. Chaieb, F. Melo and J. C. Geminard, “Experimental study of developable cones” Phys. Rev. Lett. 80, 2354 (1998).

[10] E. Cerda and L. Mahadevan, “Conical surfaces and crescent singularities in crumpled sheets” Phys. Rev. Lett. 80, 2358(1998).

[11] S. Chaieb and F. Melo, “Crescent singularities and stress focusing in a buckled thin sheet: Mechanics of developable cones”Phys. Rev. E 60, 6091 (1999).

30

[12] J. Frohlich, U. M. Studer, “Gauge-invariance and Current-Algebra in Nonrelativistic Many-Body Theory” Rev Mod Phys65, 733 (1993).

[13] M. Nelkin, “In What Sense Is Turbulence An Unsolved Problem” Science 255, 566 (1992).[14] Dearcangelis L, Hansen A, Herrmann HJ, et al., “Scaling Laws In Fracture” Phys. Rev. B 40 (1), (1989).[15] Gyure MF, Beale PD, “Dielectric-Breakdown In Continuous Models Of Metal-Loaded Dielectrics” Phys. Rev. B 46 (7),

3736 (1992).[16] A. M. Polyakov, “String Theory and Quark Confinement” Nucl Phys B Proc Sup 68, 1 (1998)[17] S. Muller, “Variational Models for Microstructure and Phase Transitions” Calculus of Variations and Geometric Evolution

Problems (Springer-Verlag, Berlin, 1999), p. 85.[18] P. Sternberg, “The effect of a singular perturbation on nonconvex variational problems” Arch. Rat. Mech. Anal. 101, 109

(1988)[19] R. V. Kohn and P. Sternberg, “Local minimizers and singular perturbations” Proc. Roy. Soc. Edinburgh A 111, 69 (1989);

I. Fonseca and L. Tartar, “The gradient theory of phase transitions for systems with two potential wells” Proc. Roy. Soc.Edinburgh A 111, 89 (1989).

[20] F. Bethuel, H. Brezis and F. Helein, “Ginzburg-Landau vortices” (Birkhauser, Boston, 1994).[21] K. Bhattacharya, R. D. James, “A theory of thin films of martensitic materials with applications to microactuators” J

Mech Phys Solids 47, 531 (1999).[22] A. E. Lobkovsky, “Boundary Layer Analysis of the Ridge Singularity in a Thin Plate” Phys Rev E 53, 3750 (1996).[23] A. Lobkovsky, S. Gentes, H. Li, D. Morse, T. A. Witten, “Scaling Properties of Stretching Ridges in a Crumpled Elastic

Sheet” Science 270, 1482 (1995).[24] M. Ortiz and G. Gioia, “The morphology and folding patterns of buckling-driven thin-film blisters” J. Mech. Phys. Solids

42, 531 (1994).[25] P. Aviles and Y. Giga, “On lower semicontinuity of a defect energy obtained by a singular limit of a Ginzburg – Landau

type energy for gradient fields” Proc. Roy. Soc. Edinburgh A 129, 1 (1999).[26] W. Jin and R. V. Kohn, “Singular Perturbation and the Energy of Folds” J. Nonlinear Sci. 10, 355 (2000).[27] B. Audoly, “Stability of straight delamination blisters”, Phys. Rev. Lett. 83, 4124 (1999).[28] A. De Simone, R. V. Kohn, S. Muller and F. Otto, “A compactness result in the gradient theory of phase transitions”,

preprint.[29] L. Ambrosio, C. De Lellis and C. Mantegazza, “Line energies for gradient fields in the plane”, preprint.[30] W. Jin and P. Sternberg, “In-plane displacements in thin film blistering”, preprint.[31] W. Jin and P. Sternberg, “Energy estimates for the von Karman Model of Thin-film Blistering”, preprint.[32] H. B. Belgacem, S. Conti, A. De Simone and S. Muller, “Rigorous bounds for the Foppl–von Karman theory of isotropically

compressed plates”, preprint.[33] E. M. Kramer, “The von Karman Equations, the Stress Function, and Elastic Ridges in High Dimensions” J. Math. Physics

38, 830 (1997).[34] E. M. Kramer, T. A. Witten, “Stress Condensation in Crushed Elastic Manifolds” Phys Rev Lett 78, 1303 (1997).[35] S. C. Venkataramani, T. A. Witten, E. M. Kramer, R. P. Geroch, “Limitations on the smooth confinement of an unstretch-

able manifold” J Math Phys 41, 5107 (2000).[36] L. D. Landau and E. M. Lifshitz, Theory of Elasticity (Pergamon Press, New York, 1959).[37] W. H. Press, S. A. Teukolsky, W. T. Vetterling and B. P. Flannery, Numerical Recipes in C (Cambridge University Press,

Cambridge, 1992).[38] M. Spivak, “A Comprehensive Introduction to Differential Geometry”, (Publish or Perish, Houston, 1979).[39] T. von Karman, The Collected Works of Theodore von Karman Butterworths Scientific Publications, London, 1956.[40] G. B. Airy, Phil. Trans. Roy. Soc. 153, 49 (1863).[41] R. M. Wald, General Relativity (The University of Chicago Press, Chicago, 1984).[42] J. C. Maxwell, Edinburgh Roy. Soc. Trans. 26 (1870), reprinted in The Scientific Papers of James Clark Maxwell (Cam-

bridge University Press, Cambridge, 1927), Vol. 2, p. 161.[43] P. Patricio and W. Krauth, Int. J. Mod. Phys. C 8, 427 (1997)[44] E. DeGiorgi, “Γ–convergenza e G–convergenza” Boll. Un. Mat. Ital. 14, 213 (1977).[45] G. Dal Maso, “An introduction to Γ – convergence”, (Birkhauser, Boston, 1993).[46] By subsheet we mean a subset of the manifold lying in a subspace of Rm.[47] The broad region of the bending energy density graph between ≈ 2% and ≈ 40% volume fraction which seems to show

clear scaling is well fit by a scaling exponent of ≈ −0.9. This suggests that the energy falls away slowly transverse tomost energetic cone generators. This observation is not contradictory to our assumption that the region of high curvaturepersists for a length of order 1/C in the transverse directions.